feat: sync upstream pi-mono (386 commits)

从上游 badlogic/pi-mono 同步 386 个提交，手动解决所有冲突：

保留 SproutClaw 独有功能：
- RPC: reload 命令、bash 流式输出、get_extensions 命令
- settings-manager: showChangelogOnStartup 设置
- agent-session: turnIndex getter
- 移除 pi 版本检查通知（interactive-mode）
- skills 系统、启动脚本、自定义工具脚本

直接采用上游版本：
- packages/ai 模型列表及测试文件
- packages/coding-agent 文档
- 其余未冲突的全部上游代码

升级 @mistralai/mistralai 至 2.2.6（修复 promptCacheKey 类型错误）

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

packages/coding-agent/test/agent-session-auto-compaction-queue.test.ts

@@ -2,7 +2,12 @@ import { existsSync, mkdirSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { Agent } from "@earendil-works/pi-agent-core";
-import { type AssistantMessage, getModel } from "@earendil-works/pi-ai";
+import {
+	type AssistantMessage,
+	createAssistantMessageEventStream,
+	fauxAssistantMessage,
+	getModel,
+} from "@earendil-works/pi-ai";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { AgentSession } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
@@ -11,51 +16,10 @@ import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 import { createTestResourceLoader } from "./utilities.ts";
 
-vi.mock("../src/core/compaction/index.js", () => ({
-	calculateContextTokens: (usage: {
-		input: number;
-		output: number;
-		cacheRead: number;
-		cacheWrite: number;
-		totalTokens?: number;
-	}) => usage.totalTokens ?? usage.input + usage.output + usage.cacheRead + usage.cacheWrite,
-	collectEntriesForBranchSummary: () => ({ entries: [], commonAncestorId: null }),
-	compact: async () => ({
-		summary: "compacted",
-		firstKeptEntryId: "entry-1",
-		tokensBefore: 100,
-		details: {},
-	}),
-	estimateContextTokens: (
-		messages: Array<{
-			role: string;
-			usage?: { input: number; output: number; cacheRead: number; cacheWrite: number; totalTokens?: number };
-			stopReason?: string;
-		}>,
-	) => {
-		// Walk backwards to find last non-error, non-aborted assistant with usage
-		for (let i = messages.length - 1; i >= 0; i--) {
-			const msg = messages[i];
-			if (msg.role === "assistant" && msg.stopReason !== "error" && msg.stopReason !== "aborted" && msg.usage) {
-				const tokens =
-					msg.usage.totalTokens ?? msg.usage.input + msg.usage.output + msg.usage.cacheRead + msg.usage.cacheWrite;
-				return { tokens, usageTokens: tokens, trailingTokens: 0, lastUsageIndex: i };
-			}
-		}
-		return { tokens: 0, usageTokens: 0, trailingTokens: 0, lastUsageIndex: null };
-	},
-	generateBranchSummary: async () => ({ summary: "", aborted: false, readFiles: [], modifiedFiles: [] }),
-	prepareCompaction: () => ({ dummy: true }),
-	shouldCompact: (
-		contextTokens: number,
-		contextWindow: number,
-		settings: { enabled: boolean; reserveTokens: number },
-	) => settings.enabled && contextTokens > contextWindow - settings.reserveTokens,
-}));
-
 describe("AgentSession auto-compaction queue resume", () => {
 	let session: AgentSession;
 	let sessionManager: SessionManager;
+	let settingsManager: SettingsManager;
 	let tempDir: string;
 
 	beforeEach(() => {
@@ -73,7 +37,7 @@ describe("AgentSession auto-compaction queue resume", () => {
 		});
 
 		sessionManager = SessionManager.inMemory();
-		const settingsManager = SettingsManager.create(tempDir, tempDir);
+		settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
 		authStorage.setRuntimeApiKey("anthropic", "test-key");
 		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
@@ -98,6 +62,57 @@ describe("AgentSession auto-compaction queue resume", () => {
 	});
 
 	it("should resume after threshold compaction when only agent-level queued messages exist", async () => {
+		settingsManager.applyOverrides({ compaction: { keepRecentTokens: 1 } });
+		const model = session.model!;
+		const now = Date.now();
+		sessionManager.appendMessage({
+			role: "user",
+			content: [{ type: "text", text: "message to compact" }],
+			timestamp: now - 1000,
+		});
+		sessionManager.appendMessage({
+			role: "assistant",
+			content: [{ type: "text", text: "assistant response to compact" }],
+			api: model.api,
+			provider: model.provider,
+			model: model.id,
+			usage: {
+				input: 100,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+				totalTokens: 100,
+				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+			},
+			stopReason: "stop",
+			timestamp: now - 500,
+		});
+		session.agent.state.messages = sessionManager.buildSessionContext().messages;
+		session.agent.streamFn = (summaryModel) => {
+			const stream = createAssistantMessageEventStream();
+			queueMicrotask(() => {
+				stream.push({
+					type: "done",
+					reason: "stop",
+					message: {
+						...fauxAssistantMessage("compacted"),
+						api: summaryModel.api,
+						provider: summaryModel.provider,
+						model: summaryModel.id,
+						usage: {
+							input: 10,
+							output: 0,
+							cacheRead: 0,
+							cacheWrite: 0,
+							totalTokens: 10,
+							cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+						},
+					},
+				});
+			});
+			return stream;
+		};
+
 		session.agent.followUp({
 			role: "custom",
 			customType: "test",

packages/coding-agent/test/agent-session-concurrent.test.ts

@@ -444,6 +444,7 @@ describe("AgentSession concurrent prompt guard", () => {
 					text: string,
 					images: unknown,
 					source: "interactive" | "rpc" | "extension",
+					streamingBehavior?: "steer" | "followUp",
 				) => Promise<{ action: "continue" }>;
 				emitBeforeAgentStart: (
 					prompt: string,
@@ -588,6 +589,7 @@ describe("AgentSession concurrent prompt guard", () => {
 					text: string,
 					images: unknown,
 					source: "interactive" | "rpc" | "extension",
+					streamingBehavior?: "steer" | "followUp",
 				) => Promise<{ action: "continue" }>;
 				emitBeforeAgentStart: (
 					prompt: string,

packages/coding-agent/test/agent-session-dynamic-tools.test.ts

@@ -72,6 +72,9 @@ describe("AgentSession dynamic tool registration", () => {
 		const readTool = allTools.find((tool) => tool.name === "read");
 
 		expect(allTools.map((tool) => tool.name)).toContain("dynamic_tool");
+		expect(dynamicTool?.promptGuidelines).toEqual([
+			"Use dynamic_tool when the user asks for dynamic behavior tests.",
+		]);
 		expect(dynamicTool?.sourceInfo).toMatchObject({
 			path: "<inline:1>",
 			source: "inline",

packages/coding-agent/test/args.test.ts

@@ -130,6 +130,11 @@ describe("parseArgs", () => {
 			expect(result.session).toBe("/path/to/session.jsonl");
 		});
 
+		test("parses --session-id", () => {
+			const result = parseArgs(["--session-id", "orchestrated-session"]);
+			expect(result.sessionId).toBe("orchestrated-session");
+		});
+
 		test("parses --fork", () => {
 			const result = parseArgs(["--fork", "1234abcd"]);
 			expect(result.fork).toBe("1234abcd");
@@ -152,6 +157,36 @@ describe("parseArgs", () => {
 		});
 	});
 
+	describe("--name flag", () => {
+		test("parses --name flag with value", () => {
+			const result = parseArgs(["--name", "my-session"]);
+			expect(result.name).toBe("my-session");
+		});
+
+		test("parses -n shorthand", () => {
+			const result = parseArgs(["-n", "quick-session"]);
+			expect(result.name).toBe("quick-session");
+		});
+
+		test("preserves empty values for main validation", () => {
+			const result = parseArgs(["--name", ""]);
+			expect(result.name).toBe("");
+		});
+
+		test("reports missing value", () => {
+			const result = parseArgs(["--name"]);
+			expect(result.diagnostics).toEqual([{ type: "error", message: "--name requires a value" }]);
+		});
+
+		test("works alongside other flags", () => {
+			const result = parseArgs(["--name", "named-run", "--print", "--model", "gpt-4o", "hello"]);
+			expect(result.name).toBe("named-run");
+			expect(result.print).toBe(true);
+			expect(result.model).toBe("gpt-4o");
+			expect(result.messages).toEqual(["hello"]);
+		});
+	});
+
 	describe("--no-session flag", () => {
 		test("parses --no-session flag", () => {
 			const result = parseArgs(["--no-session"]);
@@ -258,6 +293,28 @@ describe("parseArgs", () => {
 		});
 	});
 
+	describe("project approval flags", () => {
+		test("parses --approve", () => {
+			const result = parseArgs(["--approve"]);
+			expect(result.projectTrustOverride).toBe(true);
+		});
+
+		test("parses -a shorthand", () => {
+			const result = parseArgs(["-a"]);
+			expect(result.projectTrustOverride).toBe(true);
+		});
+
+		test("parses --no-approve", () => {
+			const result = parseArgs(["--no-approve"]);
+			expect(result.projectTrustOverride).toBe(false);
+		});
+
+		test("parses -na shorthand", () => {
+			const result = parseArgs(["-na"]);
+			expect(result.projectTrustOverride).toBe(false);
+		});
+	});
+
 	describe("--verbose flag", () => {
 		test("parses --verbose flag", () => {
 			const result = parseArgs(["--verbose"]);
@@ -303,6 +360,16 @@ describe("parseArgs", () => {
 			expect(result.tools).toEqual(["read", "bash"]);
 		});
 
+		test("parses --exclude-tools flag", () => {
+			const result = parseArgs(["--exclude-tools", "read,bash"]);
+			expect(result.excludeTools).toEqual(["read", "bash"]);
+		});
+
+		test("parses -xt shorthand", () => {
+			const result = parseArgs(["-xt", "read,bash"]);
+			expect(result.excludeTools).toEqual(["read", "bash"]);
+		});
+
 		test("parses --no-tools with explicit --tools flags", () => {
 			const result = parseArgs(["--no-tools", "--tools", "read,bash"]);
 			expect(result.noTools).toBe(true);

packages/coding-agent/test/auth-storage.test.ts

@@ -5,7 +5,8 @@ import { registerOAuthProvider } from "@earendil-works/pi-ai/oauth";
 import lockfile from "proper-lockfile";
 import { afterEach, beforeEach, describe, expect, test, vi } from "vitest";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { clearConfigValueCache } from "../src/core/resolve-config-value.ts";
+import { clearConfigValueCache, resolveConfigValueUncached } from "../src/core/resolve-config-value.ts";
+import * as shellModule from "../src/utils/shell.ts";
 
 describe("AuthStorage", () => {
 	let tempDir: string;
@@ -112,13 +113,13 @@ describe("AuthStorage", () => {
 			expect(apiKey).toBeUndefined();
 		});
 
-		test("apiKey as environment variable name resolves to env value", async () => {
+		test("apiKey with $ prefix resolves to env value", async () => {
 			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
 			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
 
 			try {
 				writeAuthJson({
-					anthropic: { type: "api_key", key: "TEST_AUTH_API_KEY_12345" },
+					anthropic: { type: "api_key", key: "$TEST_AUTH_API_KEY_12345" },
 				});
 
 				authStorage = AuthStorage.create(authJsonPath);
@@ -134,6 +135,168 @@ describe("AuthStorage", () => {
 			}
 		});
 
+		test("apiKey env bag takes precedence over process.env", async () => {
+			const originalEnv = process.env.TEST_AUTH_SCOPED_API_KEY_12345;
+			process.env.TEST_AUTH_SCOPED_API_KEY_12345 = "process-env-value";
+
+			try {
+				writeAuthJson({
+					anthropic: {
+						type: "api_key",
+						key: "$TEST_AUTH_SCOPED_API_KEY_12345",
+						env: { TEST_AUTH_SCOPED_API_KEY_12345: "credential-env-value" },
+					},
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+
+				expect(await authStorage.getApiKey("anthropic")).toBe("credential-env-value");
+				expect(authStorage.getProviderEnv("anthropic")).toEqual({
+					TEST_AUTH_SCOPED_API_KEY_12345: "credential-env-value",
+				});
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_AUTH_SCOPED_API_KEY_12345;
+				} else {
+					process.env.TEST_AUTH_SCOPED_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("apiKey with braced env syntax resolves to env value", async () => {
+			const originalEnv = process.env.TEST_AUTH_BRACED_API_KEY_12345;
+			process.env.TEST_AUTH_BRACED_API_KEY_12345 = "braced-env-api-key-value";
+			const bracedKey = "$" + "{TEST_AUTH_BRACED_API_KEY_12345}";
+
+			try {
+				writeAuthJson({
+					anthropic: { type: "api_key", key: bracedKey },
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+				const apiKey = await authStorage.getApiKey("anthropic");
+
+				expect(apiKey).toBe("braced-env-api-key-value");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_AUTH_BRACED_API_KEY_12345;
+				} else {
+					process.env.TEST_AUTH_BRACED_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("apiKey interpolates braced env references inside literals", async () => {
+			const originalPartA = process.env.TEST_AUTH_INTERPOLATED_PART_A_12345;
+			const originalPartB = process.env.TEST_AUTH_INTERPOLATED_PART_B_12345;
+			process.env.TEST_AUTH_INTERPOLATED_PART_A_12345 = "left";
+			process.env.TEST_AUTH_INTERPOLATED_PART_B_12345 = "right";
+			const interpolatedKey = [
+				"$",
+				"{TEST_AUTH_INTERPOLATED_PART_A_12345}_$",
+				"{TEST_AUTH_INTERPOLATED_PART_B_12345}",
+			].join("");
+
+			try {
+				writeAuthJson({
+					anthropic: { type: "api_key", key: interpolatedKey },
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+				const apiKey = await authStorage.getApiKey("anthropic");
+
+				expect(apiKey).toBe("left_right");
+			} finally {
+				if (originalPartA === undefined) {
+					delete process.env.TEST_AUTH_INTERPOLATED_PART_A_12345;
+				} else {
+					process.env.TEST_AUTH_INTERPOLATED_PART_A_12345 = originalPartA;
+				}
+				if (originalPartB === undefined) {
+					delete process.env.TEST_AUTH_INTERPOLATED_PART_B_12345;
+				} else {
+					process.env.TEST_AUTH_INTERPOLATED_PART_B_12345 = originalPartB;
+				}
+			}
+		});
+
+		test("apiKey with $$ prefix escapes a leading dollar", async () => {
+			writeAuthJson({
+				anthropic: { type: "api_key", key: "$$TEST_AUTH_API_KEY_12345" },
+			});
+
+			authStorage = AuthStorage.create(authJsonPath);
+			const apiKey = await authStorage.getApiKey("anthropic");
+
+			expect(apiKey).toBe("$TEST_AUTH_API_KEY_12345");
+		});
+
+		test("apiKey with $! escapes a literal bang and still interpolates later env refs", async () => {
+			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
+			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
+
+			try {
+				writeAuthJson({
+					anthropic: { type: "api_key", key: "$!literal-$TEST_AUTH_API_KEY_12345" },
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+				const apiKey = await authStorage.getApiKey("anthropic");
+
+				expect(apiKey).toBe("!literal-env-api-key-value");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_AUTH_API_KEY_12345;
+				} else {
+					process.env.TEST_AUTH_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("plain API key is used directly even when it matches an env var", async () => {
+			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
+			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
+
+			try {
+				writeAuthJson({
+					anthropic: { type: "api_key", key: "TEST_AUTH_API_KEY_12345" },
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+				const apiKey = await authStorage.getApiKey("anthropic");
+
+				expect(apiKey).toBe("TEST_AUTH_API_KEY_12345");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_AUTH_API_KEY_12345;
+				} else {
+					process.env.TEST_AUTH_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("literal public API key is not corrupted by the Windows PUBLIC env var", async () => {
+			const originalPublic = process.env.PUBLIC;
+			process.env.PUBLIC = "C:\\Users\\Public";
+
+			try {
+				writeAuthJson({
+					opencode: { type: "api_key", key: "public" },
+				});
+
+				authStorage = AuthStorage.create(authJsonPath);
+				const apiKey = await authStorage.getApiKey("opencode");
+
+				expect(apiKey).toBe("public");
+			} finally {
+				if (originalPublic === undefined) {
+					delete process.env.PUBLIC;
+				} else {
+					process.env.PUBLIC = originalPublic;
+				}
+			}
+		});
+
 		test("apiKey as literal value is used directly when not an env var", async () => {
 			// Make sure this isn't an env var
 			delete process.env.literal_api_key_value;
@@ -159,6 +322,30 @@ describe("AuthStorage", () => {
 			expect(apiKey).toBe("hello-world");
 		});
 
+		test("command config uses stdin when configured shell requires it", () => {
+			if (process.platform === "win32") return;
+			const platformDescriptor = Object.getOwnPropertyDescriptor(process, "platform");
+			vi.spyOn(shellModule, "getShellConfig").mockReturnValue({
+				shell: "/bin/bash",
+				args: ["-s"],
+				commandTransport: "stdin",
+			});
+
+			try {
+				Object.defineProperty(process, "platform", {
+					configurable: true,
+					value: "win32",
+				});
+				const nameExpansion = "$" + "{name}";
+
+				expect(resolveConfigValueUncached(`!name='World'; echo "Hello, ${nameExpansion}!"`)).toBe("Hello, World!");
+			} finally {
+				if (platformDescriptor) {
+					Object.defineProperty(process, "platform", platformDescriptor);
+				}
+			}
+		});
+
 		describe("caching", () => {
 			test("command is only executed once per process", async () => {
 				// Use a command that writes to a file to count invocations
@@ -274,7 +461,7 @@ describe("AuthStorage", () => {
 					process.env[envVarName] = "first-value";
 
 					writeAuthJson({
-						anthropic: { type: "api_key", key: envVarName },
+						anthropic: { type: "api_key", key: `$${envVarName}` },
 					});
 
 					authStorage = AuthStorage.create(authJsonPath);

packages/coding-agent/test/changelog.test.ts

@@ -0,0 +1,49 @@
+import { describe, expect, test } from "vitest";
+import { type ChangelogEntry, normalizeChangelogLinks } from "../src/utils/changelog.ts";
+
+const entry: ChangelogEntry = {
+	major: 0,
+	minor: 79,
+	patch: 0,
+	content: "",
+};
+
+describe("normalizeChangelogLinks", () => {
+	test("rewrites package-relative changelog links to tag-pinned GitHub source links", () => {
+		const markdown = [
+			"[Project Trust](README.md#project-trust)",
+			"[Extensions](docs/extensions.md#project_trust)",
+			"[Examples](examples/extensions/)",
+			"[Root README](../../README.md#supply-chain-hardening)",
+		].join("\n");
+
+		expect(normalizeChangelogLinks(markdown, entry)).toBe(
+			[
+				"[Project Trust](https://github.com/earendil-works/pi/blob/v0.79.0/packages/coding-agent/README.md#project-trust)",
+				"[Extensions](https://github.com/earendil-works/pi/blob/v0.79.0/packages/coding-agent/docs/extensions.md#project_trust)",
+				"[Examples](https://github.com/earendil-works/pi/tree/v0.79.0/packages/coding-agent/examples/extensions/)",
+				"[Root README](https://github.com/earendil-works/pi/blob/v0.79.0/README.md#supply-chain-hardening)",
+			].join("\n"),
+		);
+	});
+
+	test("canonicalizes old repository URLs without changing external links", () => {
+		const markdown = [
+			"[#5167](https://github.com/earendil-works/pi-mono/pull/5167)",
+			"[#4163](https://github.com/badlogic/pi-mono/issues/4163)",
+			"[Agent README](https://github.com/badlogic/pi-mono/blob/main/packages/agent/README.md)",
+			"[External](https://example.com/docs)",
+			"[Local anchor](#settings)",
+		].join("\n");
+
+		expect(normalizeChangelogLinks(markdown, "0.79.0")).toBe(
+			[
+				"[#5167](https://github.com/earendil-works/pi/pull/5167)",
+				"[#4163](https://github.com/earendil-works/pi/issues/4163)",
+				"[Agent README](https://github.com/earendil-works/pi/blob/v0.79.0/packages/agent/README.md)",
+				"[External](https://example.com/docs)",
+				"[Local anchor](#settings)",
+			].join("\n"),
+		);
+	});
+});

packages/coding-agent/test/clipboard-native.test.ts

@@ -0,0 +1,31 @@
+import { describe, expect, test, vi } from "vitest";
+import { type ClipboardModule, loadClipboardNative } from "../src/utils/clipboard-native.ts";
+
+type ClipboardRequire = (id: string) => unknown;
+
+const fakeClipboard: ClipboardModule = {
+	setText: async () => {},
+	hasImage: () => true,
+	getImageBinary: async () => [1, 2, 3],
+};
+
+describe("loadClipboardNative", () => {
+	test("falls back to the next require root", () => {
+		const primary = vi.fn<ClipboardRequire>(() => {
+			throw new Error("missing from bundled root");
+		});
+		const fallback = vi.fn<ClipboardRequire>(() => fakeClipboard);
+
+		expect(loadClipboardNative([primary, fallback])).toBe(fakeClipboard);
+		expect(primary).toHaveBeenCalledWith("@mariozechner/clipboard");
+		expect(fallback).toHaveBeenCalledWith("@mariozechner/clipboard");
+	});
+
+	test("returns null when no require root can load clipboard", () => {
+		const missing = vi.fn<ClipboardRequire>(() => {
+			throw new Error("missing");
+		});
+
+		expect(loadClipboardNative([missing])).toBeNull();
+	});
+});

packages/coding-agent/test/compaction-extensions.test.ts

@@ -98,6 +98,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 
 		const sessionManager = SessionManager.create(tempDir);
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
+		settingsManager.applyOverrides({ compaction: { keepRecentTokens: 1 } });
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
 		const modelRegistry = ModelRegistry.create(authStorage);
 

packages/coding-agent/test/compaction.test.ts

@@ -9,7 +9,6 @@ import {
 	calculateContextTokens,
 	compact,
 	DEFAULT_COMPACTION_SETTINGS,
-	estimateContextTokens,
 	findCutPoint,
 	getLastAssistantUsage,
 	prepareCompaction,
@@ -396,7 +395,7 @@ describe("buildSessionContext", () => {
 });
 
 describe("prepareCompaction with previous compaction", () => {
-	it("should preserve kept messages across repeated compactions when they still fit", () => {
+	it("should skip repeated compactions when kept messages still fit", () => {
 		const u1 = createMessageEntry(createUserMessage("user msg 1 (summarized by compaction1)"));
 		const a1 = createMessageEntry(createAssistantMessage("assistant msg 1"));
 		const u2 = createMessageEntry(createUserMessage("user msg 2 - kept by compaction1"));
@@ -408,29 +407,9 @@ describe("prepareCompaction with previous compaction", () => {
 		const a4 = createMessageEntry(createAssistantMessage("assistant msg 4", createMockUsage(8000, 2000)));
 
 		const pathEntries = [u1, a1, u2, a2, u3, a3, compaction1, u4, a4];
-		const contextBefore = buildSessionContext(pathEntries);
 		const preparation = prepareCompaction(pathEntries, DEFAULT_COMPACTION_SETTINGS);
 
-		expect(preparation).toBeDefined();
-		expect(preparation!.firstKeptEntryId).toBe(u2.id);
-		expect(preparation!.previousSummary).toBe("First summary");
-		expect(extractText(preparation!.messagesToSummarize)).not.toContain("First summary");
-		expect(preparation!.tokensBefore).toBe(estimateContextTokens(contextBefore.messages).tokens);
-
-		const compaction2: CompactionEntry = {
-			type: "compaction",
-			id: "compaction2-id",
-			parentId: a4.id,
-			timestamp: new Date().toISOString(),
-			summary: "Second summary",
-			firstKeptEntryId: preparation!.firstKeptEntryId,
-			tokensBefore: preparation!.tokensBefore,
-		};
-		const contextAfter = buildSessionContext([...pathEntries, compaction2]);
-		const contextAfterText = extractText(contextAfter.messages);
-
-		expect(contextAfterText).toContain("user msg 2 - kept by compaction1");
-		expect(contextAfterText).toContain("user msg 3 - kept by compaction1");
+		expect(preparation).toBeUndefined();
 	});
 
 	it("should re-summarize previously kept messages when the recent window moves past them", () => {

packages/coding-agent/test/config-value-migration.test.ts

@@ -0,0 +1,177 @@
+import * as fs from "node:fs";
+import * as os from "node:os";
+import * as path from "node:path";
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { ENV_AGENT_DIR } from "../src/config.ts";
+import { AuthStorage } from "../src/core/auth-storage.ts";
+import { ModelRegistry } from "../src/core/model-registry.ts";
+import { runMigrations } from "../src/migrations.ts";
+
+describe("config value env var syntax migration", () => {
+	const tempDirs: string[] = [];
+
+	afterEach(() => {
+		for (const dir of tempDirs.splice(0)) {
+			fs.rmSync(dir, { recursive: true, force: true });
+		}
+		vi.restoreAllMocks();
+	});
+
+	function createAgentDir(): string {
+		const agentDir = fs.mkdtempSync(path.join(os.tmpdir(), "pi-config-value-migration-test-"));
+		tempDirs.push(agentDir);
+		return agentDir;
+	}
+
+	function withAgentDir(agentDir: string, fn: () => void): void {
+		const previousAgentDir = process.env[ENV_AGENT_DIR];
+		process.env[ENV_AGENT_DIR] = agentDir;
+		try {
+			fn();
+		} finally {
+			if (previousAgentDir === undefined) {
+				delete process.env[ENV_AGENT_DIR];
+			} else {
+				process.env[ENV_AGENT_DIR] = previousAgentDir;
+			}
+		}
+	}
+
+	it("leaves uppercase auth.json API key values unchanged", () => {
+		const agentDir = createAgentDir();
+		fs.writeFileSync(
+			path.join(agentDir, "auth.json"),
+			`${JSON.stringify(
+				{
+					anthropic: { type: "api_key", key: "ANTHROPIC_API_KEY" },
+					openai: { type: "api_key", key: "$OPENAI_API_KEY" },
+					opencode: { type: "api_key", key: "public" },
+					github: { type: "oauth", access: "ACCESS_TOKEN", refresh: "REFRESH_TOKEN", expires: 1 },
+				},
+				null,
+				2,
+			)}\n`,
+			"utf-8",
+		);
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		withAgentDir(agentDir, () => runMigrations(agentDir));
+
+		const migrated = JSON.parse(fs.readFileSync(path.join(agentDir, "auth.json"), "utf-8")) as Record<
+			string,
+			Record<string, unknown>
+		>;
+		expect(migrated.anthropic.key).toBe("ANTHROPIC_API_KEY");
+		expect(migrated.openai.key).toBe("$OPENAI_API_KEY");
+		expect(migrated.opencode.key).toBe("public");
+		expect(migrated.github.access).toBe("ACCESS_TOKEN");
+		expect(logSpy).not.toHaveBeenCalled();
+	});
+
+	it.each([
+		["malformed", '{\n  "providers": {\n'],
+		["blank", ""],
+	])("does not throw on %s models.json during migrations", (_name, content) => {
+		const agentDir = createAgentDir();
+		const modelsPath = path.join(agentDir, "models.json");
+		fs.writeFileSync(modelsPath, content, "utf-8");
+
+		withAgentDir(agentDir, () => expect(() => runMigrations(agentDir)).not.toThrow());
+
+		expect(fs.readFileSync(modelsPath, "utf-8")).toBe(content);
+		const registry = ModelRegistry.create(AuthStorage.create(path.join(agentDir, "auth.json")), modelsPath);
+		const loadError = registry.getError();
+		expect(loadError).toContain("Failed to parse models.json");
+		expect(loadError).toContain(`File: ${modelsPath}`);
+	});
+
+	it("leaves uppercase models.json API key and header values unchanged", async () => {
+		const agentDir = createAgentDir();
+		const envKeys = ["CUSTOM_API_KEY", "HEADER_API_KEY", "MODEL_API_KEY", "OVERRIDE_API_KEY"];
+		const savedEnv: Record<string, string | undefined> = {};
+		for (const key of envKeys) {
+			savedEnv[key] = process.env[key];
+			process.env[key] = `env-${key}`;
+		}
+
+		try {
+			fs.writeFileSync(
+				path.join(agentDir, "models.json"),
+				`${JSON.stringify(
+					{
+						providers: {
+							"custom-provider": {
+								baseUrl: "https://example.com/v1",
+								apiKey: "CUSTOM_API_KEY",
+								api: "openai-completions",
+								headers: {
+									"x-api-key": "HEADER_API_KEY",
+									"x-literal": "literal",
+								},
+								models: [
+									{
+										id: "model-a",
+										headers: { "x-model-key": "MODEL_API_KEY" },
+									},
+								],
+								modelOverrides: {
+									"model-b": { headers: { "x-override-key": "OVERRIDE_API_KEY" } },
+								},
+							},
+						},
+					},
+					null,
+					2,
+				)}\n`,
+				"utf-8",
+			);
+			const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+			withAgentDir(agentDir, () => runMigrations(agentDir));
+
+			const migrated = JSON.parse(fs.readFileSync(path.join(agentDir, "models.json"), "utf-8")) as {
+				providers: Record<
+					string,
+					{
+						apiKey?: string;
+						headers?: Record<string, string>;
+						models?: Array<{ headers?: Record<string, string> }>;
+						modelOverrides?: Record<string, { headers?: Record<string, string> }>;
+					}
+				>;
+			};
+			const provider = migrated.providers["custom-provider"]!;
+			expect(provider.apiKey).toBe("CUSTOM_API_KEY");
+			expect(provider.headers?.["x-api-key"]).toBe("HEADER_API_KEY");
+			expect(provider.headers?.["x-literal"]).toBe("literal");
+			expect(provider.models?.[0]?.headers?.["x-model-key"]).toBe("MODEL_API_KEY");
+			expect(provider.modelOverrides?.["model-b"]?.headers?.["x-override-key"]).toBe("OVERRIDE_API_KEY");
+			expect(logSpy).not.toHaveBeenCalled();
+
+			const registry = ModelRegistry.create(
+				AuthStorage.create(path.join(agentDir, "auth.json")),
+				path.join(agentDir, "models.json"),
+			);
+			const model = registry.find("custom-provider", "model-a");
+			expect(model).toBeDefined();
+			expect(await registry.getApiKeyForProvider("custom-provider")).toBe("CUSTOM_API_KEY");
+			expect(await registry.getApiKeyAndHeaders(model!)).toMatchObject({
+				ok: true,
+				apiKey: "CUSTOM_API_KEY",
+				headers: {
+					"x-api-key": "HEADER_API_KEY",
+					"x-literal": "literal",
+					"x-model-key": "MODEL_API_KEY",
+				},
+			});
+		} finally {
+			for (const key of envKeys) {
+				if (savedEnv[key] === undefined) {
+					delete process.env[key];
+				} else {
+					process.env[key] = savedEnv[key];
+				}
+			}
+		}
+	});
+});

packages/coding-agent/test/config.test.ts

@@ -153,7 +153,7 @@ describe("detectInstallMethod", () => {
 
 		expect(detectInstallMethod()).toBe("pnpm");
 		expect(getUpdateInstruction("@earendil-works/pi-coding-agent")).toBe(
-			"Run: pnpm install -g --ignore-scripts @earendil-works/pi-coding-agent",
+			"Run: pnpm install -g --ignore-scripts --config.minimumReleaseAge=0 @earendil-works/pi-coding-agent",
 		);
 	});
 
@@ -175,8 +175,16 @@ describe("detectInstallMethod", () => {
 		expect(detectInstallMethod()).toBe("npm");
 		expect(command).toEqual({
 			command: "npm",
-			args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "@earendil-works/pi-coding-agent"],
-			display: `npm --prefix ${prefix} install -g --ignore-scripts @earendil-works/pi-coding-agent`,
+			args: [
+				"--prefix",
+				prefix,
+				"install",
+				"-g",
+				"--ignore-scripts",
+				"--min-release-age=0",
+				"@earendil-works/pi-coding-agent",
+			],
+			display: `npm --prefix ${prefix} install -g --ignore-scripts --min-release-age=0 @earendil-works/pi-coding-agent`,
 		});
 	});
 
@@ -187,8 +195,8 @@ describe("detectInstallMethod", () => {
 
 		expect(command).toEqual({
 			command: "npm",
-			args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "@new-scope/pi"],
-			display: `npm --prefix ${prefix} uninstall -g @mariozechner/pi-coding-agent && npm --prefix ${prefix} install -g --ignore-scripts @new-scope/pi`,
+			args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "--min-release-age=0", "@new-scope/pi"],
+			display: `npm --prefix ${prefix} uninstall -g @mariozechner/pi-coding-agent && npm --prefix ${prefix} install -g --ignore-scripts --min-release-age=0 @new-scope/pi`,
 			steps: [
 				{
 					command: "npm",
@@ -197,8 +205,8 @@ describe("detectInstallMethod", () => {
 				},
 				{
 					command: "npm",
-					args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "@new-scope/pi"],
-					display: `npm --prefix ${prefix} install -g --ignore-scripts @new-scope/pi`,
+					args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "--min-release-age=0", "@new-scope/pi"],
+					display: `npm --prefix ${prefix} install -g --ignore-scripts --min-release-age=0 @new-scope/pi`,
 				},
 			],
 		});
@@ -211,8 +219,16 @@ describe("detectInstallMethod", () => {
 
 		expect(command).toEqual({
 			command: "npm",
-			args: ["--prefix", prefix, "install", "-g", "--ignore-scripts", "@earendil-works/pi-coding-agent"],
-			display: `npm --prefix ${prefix} install -g --ignore-scripts @earendil-works/pi-coding-agent`,
+			args: [
+				"--prefix",
+				prefix,
+				"install",
+				"-g",
+				"--ignore-scripts",
+				"--min-release-age=0",
+				"@earendil-works/pi-coding-agent",
+			],
+			display: `npm --prefix ${prefix} install -g --ignore-scripts --min-release-age=0 @earendil-works/pi-coding-agent`,
 		});
 	});
 
@@ -227,6 +243,7 @@ describe("detectInstallMethod", () => {
 			"install",
 			"-g",
 			"--ignore-scripts",
+			"--min-release-age=0",
 			"@earendil-works/pi-coding-agent",
 		]);
 	});
@@ -237,7 +254,7 @@ describe("detectInstallMethod", () => {
 		const command = getSelfUpdateCommand("@earendil-works/pi-coding-agent");
 
 		expect(command?.display).toBe(
-			`npm --prefix "${prefix}" install -g --ignore-scripts @earendil-works/pi-coding-agent`,
+			`npm --prefix "${prefix}" install -g --ignore-scripts --min-release-age=0 @earendil-works/pi-coding-agent`,
 		);
 	});
 
@@ -248,7 +265,7 @@ describe("detectInstallMethod", () => {
 
 		expect(detectInstallMethod()).toBe("npm");
 		expect(getUpdateInstruction("@earendil-works/pi-coding-agent")).toBe(
-			"Run: npm install -g --ignore-scripts @earendil-works/pi-coding-agent",
+			"Run: npm install -g --ignore-scripts --min-release-age=0 @earendil-works/pi-coding-agent",
 		);
 	});
 
@@ -260,8 +277,8 @@ describe("detectInstallMethod", () => {
 		expect(detectInstallMethod()).toBe("bun");
 		expect(command).toEqual({
 			command: "bun",
-			args: ["install", "-g", "--ignore-scripts", "@earendil-works/pi-coding-agent"],
-			display: "bun install -g --ignore-scripts @earendil-works/pi-coding-agent",
+			args: ["install", "-g", "--ignore-scripts", "--minimum-release-age=0", "@earendil-works/pi-coding-agent"],
+			display: "bun install -g --ignore-scripts --minimum-release-age=0 @earendil-works/pi-coding-agent",
 		});
 	});
 
@@ -273,8 +290,9 @@ describe("detectInstallMethod", () => {
 		expect(detectInstallMethod()).toBe("pnpm");
 		expect(command).toEqual({
 			command: "pnpm",
-			args: ["install", "-g", "--ignore-scripts", "@new-scope/pi"],
-			display: "pnpm remove -g @mariozechner/pi-coding-agent && pnpm install -g --ignore-scripts @new-scope/pi",
+			args: ["install", "-g", "--ignore-scripts", "--config.minimumReleaseAge=0", "@new-scope/pi"],
+			display:
+				"pnpm remove -g @mariozechner/pi-coding-agent && pnpm install -g --ignore-scripts --config.minimumReleaseAge=0 @new-scope/pi",
 			steps: [
 				{
 					command: "pnpm",
@@ -283,8 +301,8 @@ describe("detectInstallMethod", () => {
 				},
 				{
 					command: "pnpm",
-					args: ["install", "-g", "--ignore-scripts", "@new-scope/pi"],
-					display: "pnpm install -g --ignore-scripts @new-scope/pi",
+					args: ["install", "-g", "--ignore-scripts", "--config.minimumReleaseAge=0", "@new-scope/pi"],
+					display: "pnpm install -g --ignore-scripts --config.minimumReleaseAge=0 @new-scope/pi",
 				},
 			],
 		});
@@ -328,8 +346,8 @@ describe("detectInstallMethod", () => {
 		expect(detectInstallMethod()).toBe("pnpm");
 		expect(command).toEqual({
 			command: "pnpm",
-			args: ["install", "-g", "--ignore-scripts", packageName],
-			display: `pnpm install -g --ignore-scripts ${packageName}`,
+			args: ["install", "-g", "--ignore-scripts", "--config.minimumReleaseAge=0", packageName],
+			display: `pnpm install -g --ignore-scripts --config.minimumReleaseAge=0 ${packageName}`,
 		});
 	});
 
@@ -366,8 +384,9 @@ describe("detectInstallMethod", () => {
 		expect(detectInstallMethod()).toBe("bun");
 		expect(command).toEqual({
 			command: "bun",
-			args: ["install", "-g", "--ignore-scripts", "@new-scope/pi"],
-			display: "bun uninstall -g @mariozechner/pi-coding-agent && bun install -g --ignore-scripts @new-scope/pi",
+			args: ["install", "-g", "--ignore-scripts", "--minimum-release-age=0", "@new-scope/pi"],
+			display:
+				"bun uninstall -g @mariozechner/pi-coding-agent && bun install -g --ignore-scripts --minimum-release-age=0 @new-scope/pi",
 			steps: [
 				{
 					command: "bun",
@@ -376,8 +395,8 @@ describe("detectInstallMethod", () => {
 				},
 				{
 					command: "bun",
-					args: ["install", "-g", "--ignore-scripts", "@new-scope/pi"],
-					display: "bun install -g --ignore-scripts @new-scope/pi",
+					args: ["install", "-g", "--ignore-scripts", "--minimum-release-age=0", "@new-scope/pi"],
+					display: "bun install -g --ignore-scripts --minimum-release-age=0 @new-scope/pi",
 				},
 			],
 		});

packages/coding-agent/test/experimental.test.ts

@@ -0,0 +1,44 @@
+import { afterEach, describe, expect, it } from "vitest";
+import { areExperimentalFeaturesEnabled } from "../src/core/experimental.ts";
+
+describe("areExperimentalFeaturesEnabled", () => {
+	const originalPiExperimental = process.env.PI_EXPERIMENTAL;
+
+	afterEach(() => {
+		if (originalPiExperimental === undefined) {
+			delete process.env.PI_EXPERIMENTAL;
+		} else {
+			process.env.PI_EXPERIMENTAL = originalPiExperimental;
+		}
+	});
+
+	it("returns false when PI_EXPERIMENTAL is unset", () => {
+		delete process.env.PI_EXPERIMENTAL;
+
+		expect(areExperimentalFeaturesEnabled()).toBe(false);
+	});
+
+	it("returns false when PI_EXPERIMENTAL is empty", () => {
+		process.env.PI_EXPERIMENTAL = "";
+
+		expect(areExperimentalFeaturesEnabled()).toBe(false);
+	});
+
+	it("returns true when PI_EXPERIMENTAL is set to 1", () => {
+		process.env.PI_EXPERIMENTAL = "1";
+
+		expect(areExperimentalFeaturesEnabled()).toBe(true);
+	});
+
+	it("returns false when PI_EXPERIMENTAL is set to 0", () => {
+		process.env.PI_EXPERIMENTAL = "0";
+
+		expect(areExperimentalFeaturesEnabled()).toBe(false);
+	});
+
+	it("returns false when PI_EXPERIMENTAL is set to a non-1 value", () => {
+		process.env.PI_EXPERIMENTAL = "true";
+
+		expect(areExperimentalFeaturesEnabled()).toBe(false);
+	});
+});

packages/coding-agent/test/export-html-xss.test.ts

@@ -4,15 +4,20 @@ import { describe, expect, it } from "vitest";
 describe("export HTML markdown link sanitization", () => {
 	const templateJs = readFileSync(new URL("../src/core/export-html/template.js", import.meta.url), "utf-8");
 
-	it("overrides the marked link renderer to block javascript: protocol", () => {
-		// The custom link renderer must check for dangerous protocols
+	it("overrides the marked link renderer to use scheme allow-list sanitization", () => {
 		expect(templateJs).toMatch(/link\s*\(\s*token\s*\)/);
-		expect(templateJs).toMatch(/javascript/i);
-		expect(templateJs).toMatch(/vbscript/i);
+		expect(templateJs).toMatch(/sanitizeMarkdownUrl\(token\.href\)/);
+		expect(templateJs).toMatch(/\^\(https\?\|mailto\|tel\|ftp\)/);
 	});
 
-	it("overrides the marked image renderer to block javascript: protocol", () => {
+	it("overrides the marked image renderer to use scheme allow-list sanitization", () => {
 		expect(templateJs).toMatch(/image\s*\(\s*token\s*\)/);
+		expect(templateJs).toMatch(/sanitizeMarkdownUrl\(token\.href\)/);
+	});
+
+	it("strips C0 controls before checking and emitting markdown URLs", () => {
+		expect(templateJs).toContain("replace(/[\\x00-\\x1f\\x7f]/g, '')");
+		expect(templateJs).not.toMatch(/\^\\s\*\(javascript\|vbscript\|data\):/i);
 	});
 
 	it("escapes href attributes in the custom link renderer", () => {

packages/coding-agent/test/extensions-input-event.test.ts

@@ -94,6 +94,18 @@ describe("Input Event", () => {
 		}
 	});
 
+	it("passes streamingBehavior correctly", async () => {
+		const r = await createRunner(
+			`export default p => p.on("input", async e => { globalThis.testVar = e.streamingBehavior; return { action: "continue" }; });`,
+		);
+		await r.emitInput("x", undefined, "interactive", "steer");
+		expect((globalThis as any).testVar).toBe("steer");
+		await r.emitInput("x", undefined, "interactive", "followUp");
+		expect((globalThis as any).testVar).toBe("followUp");
+		await r.emitInput("x", undefined, "interactive");
+		expect((globalThis as any).testVar).toBeUndefined();
+	});
+
 	it("catches handler errors and continues", async () => {
 		const r = await createRunner(`export default p => p.on("input", async () => { throw new Error("boom"); });`);
 		const errs: string[] = [];

packages/coding-agent/test/extensions-runner.test.ts

@@ -7,9 +7,14 @@ import * as os from "node:os";
 import * as path from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { createExtensionRuntime, discoverAndLoadExtensions } from "../src/core/extensions/loader.ts";
-import { ExtensionRunner } from "../src/core/extensions/runner.ts";
-import type { ExtensionActions, ExtensionContextActions, ProviderConfig } from "../src/core/extensions/types.ts";
+import { createExtensionRuntime, discoverAndLoadExtensions, loadExtensions } from "../src/core/extensions/loader.ts";
+import { ExtensionRunner, emitProjectTrustEvent } from "../src/core/extensions/runner.ts";
+import type {
+	ExtensionActions,
+	ExtensionContextActions,
+	ExtensionUIContext,
+	ProviderConfig,
+} from "../src/core/extensions/types.ts";
 import { KeybindingsManager, type KeyId } from "../src/core/keybindings.ts";
 import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
@@ -36,7 +41,7 @@ describe("ExtensionRunner", () => {
 
 	const providerModelConfig: ProviderConfig = {
 		baseUrl: "https://provider.test/v1",
-		apiKey: "PROVIDER_TEST_KEY",
+		apiKey: "provider-test-key",
 		api: "openai-completions",
 		models: [
 			{
@@ -71,6 +76,7 @@ describe("ExtensionRunner", () => {
 	const extensionContextActions: ExtensionContextActions = {
 		getModel: () => undefined,
 		isIdle: () => true,
+		isProjectTrusted: () => true,
 		getSignal: () => undefined,
 		abort: () => {},
 		hasPendingMessages: () => false,
@@ -80,6 +86,45 @@ describe("ExtensionRunner", () => {
 		getSystemPrompt: () => "",
 	};
 
+	describe("project_trust", () => {
+		it("continues past undecided handlers and returns the first yes/no decision", async () => {
+			const undecidedPath = path.join(extensionsDir, "undecided.ts");
+			const decidedPath = path.join(extensionsDir, "decided.ts");
+			fs.writeFileSync(
+				undecidedPath,
+				`export default function(pi) {
+	pi.on("project_trust", () => ({ trusted: "undecided", remember: true }));
+}`,
+			);
+			fs.writeFileSync(
+				decidedPath,
+				`export default function(pi) {
+	pi.on("project_trust", () => ({ trusted: "no", remember: true }));
+}`,
+			);
+
+			const extensionsResult = await loadExtensions([undecidedPath, decidedPath], tempDir);
+			const result = await emitProjectTrustEvent(
+				extensionsResult,
+				{ type: "project_trust", cwd: tempDir },
+				{
+					cwd: tempDir,
+					mode: "tui",
+					hasUI: false,
+					ui: {
+						select: async () => undefined,
+						confirm: async () => false,
+						input: async () => undefined,
+						notify: () => {},
+					},
+				},
+			);
+
+			expect(result.result).toEqual({ trusted: "no", remember: true });
+			expect(result.errors).toEqual([]);
+		});
+	});
+
 	describe("shortcut conflicts", () => {
 		it("warns when extension shortcut conflicts with built-in", async () => {
 			const extCode = `
@@ -441,6 +486,50 @@ describe("ExtensionRunner", () => {
 			controller.abort();
 			expect(ctx.signal?.aborted).toBe(true);
 		});
+
+		it("exposes print mode and hasUI false by default", async () => {
+			const result = await discoverAndLoadExtensions([], tempDir, tempDir);
+			const runner = new ExtensionRunner(result.extensions, result.runtime, tempDir, sessionManager, modelRegistry);
+			runner.bindCore(extensionActions, extensionContextActions);
+
+			const ctx = runner.createContext();
+			expect(ctx.mode).toBe("print");
+			expect(ctx.hasUI).toBe(false);
+		});
+
+		it("exposes project trust state on ExtensionContext", async () => {
+			const result = await discoverAndLoadExtensions([], tempDir, tempDir);
+			const runner = new ExtensionRunner(result.extensions, result.runtime, tempDir, sessionManager, modelRegistry);
+			runner.bindCore(extensionActions, {
+				...extensionContextActions,
+				isProjectTrusted: () => false,
+			});
+
+			const ctx = runner.createContext();
+			expect(ctx.isProjectTrusted()).toBe(false);
+		});
+
+		it("exposes rpc mode with hasUI true when an RPC UI context is provided", async () => {
+			const result = await discoverAndLoadExtensions([], tempDir, tempDir);
+			const runner = new ExtensionRunner(result.extensions, result.runtime, tempDir, sessionManager, modelRegistry);
+			runner.bindCore(extensionActions, extensionContextActions);
+			runner.setUIContext({} as ExtensionUIContext, "rpc");
+
+			const ctx = runner.createContext();
+			expect(ctx.mode).toBe("rpc");
+			expect(ctx.hasUI).toBe(true);
+		});
+
+		it("exposes tui mode with hasUI true when a TUI UI context is provided", async () => {
+			const result = await discoverAndLoadExtensions([], tempDir, tempDir);
+			const runner = new ExtensionRunner(result.extensions, result.runtime, tempDir, sessionManager, modelRegistry);
+			runner.bindCore(extensionActions, extensionContextActions);
+			runner.setUIContext({} as ExtensionUIContext, "tui");
+
+			const ctx = runner.createContext();
+			expect(ctx.mode).toBe("tui");
+			expect(ctx.hasUI).toBe(true);
+		});
 	});
 
 	describe("error handling", () => {

packages/coding-agent/test/file-mutation-queue.test.ts

@@ -10,6 +10,18 @@ function delay(ms: number): Promise<void> {
 	return new Promise((resolve) => setTimeout(resolve, ms));
 }
 
+function createDeferred(): { promise: Promise<void>; resolve: () => void } {
+	let resolve!: () => void;
+	const promise = new Promise<void>((promiseResolve) => {
+		resolve = promiseResolve;
+	});
+	return { promise, resolve };
+}
+
+async function resolvesWithin(promise: Promise<unknown>, ms: number): Promise<boolean> {
+	return Promise.race([promise.then(() => true), delay(ms).then(() => false)]);
+}
+
 const tempDirs: string[] = [];
 
 async function createTempDir(): Promise<string> {
@@ -160,4 +172,103 @@ describe("built-in edit and write tools", () => {
 		const content = await readFile(filePath, "utf8");
 		expect(content).toBe("replacement\n");
 	});
+
+	it("keeps write queue locked while an aborted write is still in flight", async () => {
+		const dir = await createTempDir();
+		const filePath = join(dir, "abort-write.txt");
+		const firstWriteStarted = createDeferred();
+		const finishFirstWrite = createDeferred();
+		const secondWriteStarted = createDeferred();
+		let firstWriteSettled = false;
+
+		const writeTool = createWriteTool(dir, {
+			operations: {
+				mkdir: async () => {},
+				writeFile: async (path, content) => {
+					if (content === "first\n") {
+						firstWriteStarted.resolve();
+						await finishFirstWrite.promise;
+						await writeFile(path, content, "utf8");
+						firstWriteSettled = true;
+						return;
+					}
+
+					if (content === "second\n") {
+						expect(firstWriteSettled).toBe(true);
+						secondWriteStarted.resolve();
+					}
+					await writeFile(path, content, "utf8");
+				},
+			},
+		});
+
+		const controller = new AbortController();
+		const firstWrite = writeTool.execute("call-1", { path: filePath, content: "first\n" }, controller.signal);
+		await firstWriteStarted.promise;
+		controller.abort();
+
+		const secondWrite = writeTool.execute("call-2", { path: filePath, content: "second\n" });
+		expect(await resolvesWithin(secondWriteStarted.promise, 20)).toBe(false);
+
+		finishFirstWrite.resolve();
+		await expect(firstWrite).rejects.toThrow("Operation aborted");
+		await secondWrite;
+
+		const content = await readFile(filePath, "utf8");
+		expect(content).toBe("second\n");
+	});
+
+	it("keeps edit queue locked while an aborted edit write is still in flight", async () => {
+		const dir = await createTempDir();
+		const filePath = join(dir, "abort-edit.txt");
+		await writeFile(filePath, "alpha\nbeta\n", "utf8");
+		const firstWriteStarted = createDeferred();
+		const finishFirstWrite = createDeferred();
+		const secondWriteStarted = createDeferred();
+		let firstWriteSettled = false;
+
+		const editTool = createEditTool(dir, {
+			operations: {
+				access,
+				readFile,
+				writeFile: async (path, content) => {
+					if (content === "ALPHA\nbeta\n") {
+						firstWriteStarted.resolve();
+						await finishFirstWrite.promise;
+						await writeFile(path, content, "utf8");
+						firstWriteSettled = true;
+						return;
+					}
+
+					if (content === "ALPHA\nBETA\n" || content === "alpha\nBETA\n") {
+						expect(firstWriteSettled).toBe(true);
+						secondWriteStarted.resolve();
+					}
+					await writeFile(path, content, "utf8");
+				},
+			},
+		});
+
+		const controller = new AbortController();
+		const firstEdit = editTool.execute(
+			"call-1",
+			{ path: filePath, edits: [{ oldText: "alpha", newText: "ALPHA" }] },
+			controller.signal,
+		);
+		await firstWriteStarted.promise;
+		controller.abort();
+
+		const secondEdit = editTool.execute("call-2", {
+			path: filePath,
+			edits: [{ oldText: "beta", newText: "BETA" }],
+		});
+		expect(await resolvesWithin(secondWriteStarted.promise, 20)).toBe(false);
+
+		finishFirstWrite.resolve();
+		await expect(firstEdit).rejects.toThrow("Operation aborted");
+		await secondEdit;
+
+		const content = await readFile(filePath, "utf8");
+		expect(content).toBe("ALPHA\nBETA\n");
+	});
 });

packages/coding-agent/test/first-time-setup-fork.test.ts

@@ -0,0 +1,39 @@
+import { mkdtempSync, rmSync } from "fs";
+import { tmpdir } from "os";
+import { join } from "path";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+vi.mock("../src/config.ts", async (importOriginal) => {
+	const actual = await importOriginal();
+	return {
+		...(actual as Record<string, unknown>),
+		PACKAGE_NAME: "@example/pi-coding-agent",
+	};
+});
+
+import { shouldRunFirstTimeSetup } from "../src/cli/startup-ui.ts";
+
+describe("shouldRunFirstTimeSetup in forked distributions", () => {
+	const originalPiExperimental = process.env.PI_EXPERIMENTAL;
+	let tempDir: string;
+	let settingsPath: string;
+
+	beforeEach(() => {
+		tempDir = mkdtempSync(join(tmpdir(), "pi-first-time-setup-fork-"));
+		settingsPath = join(tempDir, "settings.json");
+		process.env.PI_EXPERIMENTAL = "1";
+	});
+
+	afterEach(() => {
+		rmSync(tempDir, { recursive: true, force: true });
+		if (originalPiExperimental === undefined) {
+			delete process.env.PI_EXPERIMENTAL;
+		} else {
+			process.env.PI_EXPERIMENTAL = originalPiExperimental;
+		}
+	});
+
+	it("returns false for a forked package", () => {
+		expect(shouldRunFirstTimeSetup(settingsPath)).toBe(false);
+	});
+});

packages/coding-agent/test/first-time-setup.test.ts

@@ -0,0 +1,95 @@
+import { mkdtempSync, rmSync, writeFileSync } from "fs";
+import { tmpdir } from "os";
+import { join } from "path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { shouldRunFirstTimeSetup } from "../src/cli/startup-ui.ts";
+import { ENV_AGENT_DIR } from "../src/config.ts";
+import { SettingsManager } from "../src/core/settings-manager.ts";
+
+describe("shouldRunFirstTimeSetup", () => {
+	const originalPiExperimental = process.env.PI_EXPERIMENTAL;
+	const originalAgentDir = process.env[ENV_AGENT_DIR];
+	let tempDir: string;
+	let settingsPath: string;
+
+	beforeEach(() => {
+		tempDir = mkdtempSync(join(tmpdir(), "pi-first-time-setup-"));
+		settingsPath = join(tempDir, "settings.json");
+		process.env.PI_EXPERIMENTAL = "1";
+		delete process.env[ENV_AGENT_DIR];
+	});
+
+	afterEach(() => {
+		rmSync(tempDir, { recursive: true, force: true });
+		if (originalPiExperimental === undefined) {
+			delete process.env.PI_EXPERIMENTAL;
+		} else {
+			process.env.PI_EXPERIMENTAL = originalPiExperimental;
+		}
+		if (originalAgentDir === undefined) {
+			delete process.env[ENV_AGENT_DIR];
+		} else {
+			process.env[ENV_AGENT_DIR] = originalAgentDir;
+		}
+	});
+
+	it("returns true when experimental, default agent dir, and no settings.json", () => {
+		expect(shouldRunFirstTimeSetup(settingsPath)).toBe(true);
+	});
+
+	it("returns false when experimental features are disabled", () => {
+		delete process.env.PI_EXPERIMENTAL;
+
+		expect(shouldRunFirstTimeSetup(settingsPath)).toBe(false);
+	});
+
+	it("returns false when a custom agent dir is set", () => {
+		process.env[ENV_AGENT_DIR] = tempDir;
+
+		expect(shouldRunFirstTimeSetup(settingsPath)).toBe(false);
+	});
+
+	it("returns false when settings.json already exists", () => {
+		writeFileSync(settingsPath, "{}", "utf-8");
+
+		expect(shouldRunFirstTimeSetup(settingsPath)).toBe(false);
+	});
+});
+
+describe("analytics settings", () => {
+	it("defaults to disabled with no tracking identifier", () => {
+		const manager = SettingsManager.inMemory();
+
+		expect(manager.getEnableAnalytics()).toBe(false);
+		expect(manager.getTrackingId()).toBeUndefined();
+	});
+
+	it("generates a tracking identifier on opt-in", () => {
+		const manager = SettingsManager.inMemory();
+
+		manager.setEnableAnalytics(true);
+
+		expect(manager.getEnableAnalytics()).toBe(true);
+		expect(manager.getTrackingId()).toMatch(/^[0-9a-f-]{36}$/);
+	});
+
+	it("does not generate a tracking identifier on opt-out", () => {
+		const manager = SettingsManager.inMemory();
+
+		manager.setEnableAnalytics(false);
+
+		expect(manager.getEnableAnalytics()).toBe(false);
+		expect(manager.getTrackingId()).toBeUndefined();
+	});
+
+	it("keeps the tracking identifier when toggling analytics", () => {
+		const manager = SettingsManager.inMemory();
+
+		manager.setEnableAnalytics(true);
+		const trackingId = manager.getTrackingId();
+		manager.setEnableAnalytics(false);
+		manager.setEnableAnalytics(true);
+
+		expect(manager.getTrackingId()).toBe(trackingId);
+	});
+});

packages/coding-agent/test/footer-width.test.ts

@@ -2,8 +2,9 @@ import { visibleWidth } from "@earendil-works/pi-tui";
 import { beforeAll, describe, expect, it } from "vitest";
 import type { AgentSession } from "../src/core/agent-session.ts";
 import type { ReadonlyFooterDataProvider } from "../src/core/footer-data-provider.ts";
-import { FooterComponent } from "../src/modes/interactive/components/footer.ts";
+import { FooterComponent, formatCwdForFooter } from "../src/modes/interactive/components/footer.ts";
 import { initTheme } from "../src/modes/interactive/theme/theme.ts";
+import { stripAnsi } from "../src/utils/ansi.ts";
 
 type AssistantUsage = {
 	input: number;
@@ -73,6 +74,17 @@ function createFooterData(providerCount: number): ReadonlyFooterDataProvider {
 	return provider;
 }
 
+describe("formatCwdForFooter", () => {
+	it("does not abbreviate sibling paths that share the home prefix", () => {
+		expect(formatCwdForFooter("/home/user2", "/home/user")).toBe("/home/user2");
+	});
+
+	it("abbreviates the home directory and descendants", () => {
+		expect(formatCwdForFooter("/home/user", "/home/user")).toBe("~");
+		expect(formatCwdForFooter("/home/user/project", "/home/user")).toBe("~/project");
+	});
+});
+
 describe("FooterComponent width handling", () => {
 	beforeAll(() => {
 		initTheme(undefined, false);
@@ -112,4 +124,21 @@ describe("FooterComponent width handling", () => {
 			expect(visibleWidth(line)).toBeLessThanOrEqual(width);
 		}
 	});
+
+	it("shows the latest cache hit rate when cache usage is present", () => {
+		const session = createSession({
+			sessionName: "",
+			usage: {
+				input: 100,
+				output: 10,
+				cacheRead: 50,
+				cacheWrite: 50,
+				cost: { total: 0.001 },
+			},
+		});
+		const footer = new FooterComponent(session, createFooterData(1));
+
+		const statsLine = stripAnsi(footer.render(120)[1]);
+		expect(statsLine).toContain("CH25.0%");
+	});
 });

packages/coding-agent/test/format-resume-command.test.ts

@@ -0,0 +1,135 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { APP_NAME } from "../src/config.ts";
+import type { SessionManager } from "../src/core/session-manager.ts";
+import { formatResumeCommand } from "../src/modes/interactive/interactive-mode.ts";
+
+const tempDirs: string[] = [];
+const originalStdoutIsTTY = Object.getOwnPropertyDescriptor(process.stdout, "isTTY");
+
+afterEach(() => {
+	if (originalStdoutIsTTY) {
+		Object.defineProperty(process.stdout, "isTTY", originalStdoutIsTTY);
+	} else {
+		Reflect.deleteProperty(process.stdout, "isTTY");
+	}
+
+	for (const dir of tempDirs.splice(0)) {
+		rmSync(dir, { recursive: true, force: true });
+	}
+});
+
+function setStdoutIsTTY(value: boolean): void {
+	Object.defineProperty(process.stdout, "isTTY", { configurable: true, value });
+}
+
+function createTempFile(): string {
+	const dir = mkdtempSync(join(tmpdir(), "pi-format-resume-command-"));
+	tempDirs.push(dir);
+	const file = join(dir, "session.jsonl");
+	writeFileSync(file, "\n");
+	return file;
+}
+
+function createSessionManager(options: {
+	persisted?: boolean;
+	sessionFile?: string;
+	sessionId?: string;
+	sessionDir?: string;
+	usesDefaultSessionDir?: boolean;
+}): SessionManager {
+	return {
+		isPersisted: () => options.persisted ?? true,
+		getSessionFile: () => options.sessionFile,
+		getSessionId: () => options.sessionId ?? "0197f6e4-4cf9-7f44-a2d8-f8f7f49ee9d3",
+		getSessionDir: () => options.sessionDir ?? "/tmp/pi-sessions",
+		usesDefaultSessionDir: () => options.usesDefaultSessionDir ?? true,
+	} as unknown as SessionManager;
+}
+
+describe("formatResumeCommand", () => {
+	it("returns a session resume command for default session dirs", () => {
+		setStdoutIsTTY(true);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({ sessionFile, sessionId: "test-session" });
+
+		expect(formatResumeCommand(sessionManager)).toBe(`${APP_NAME} --session test-session`);
+	});
+
+	it("includes unquoted safe session dirs for non-default session dirs", () => {
+		setStdoutIsTTY(true);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({
+			sessionFile,
+			sessionId: "test-session",
+			sessionDir: "/tmp/custom-pi-sessions",
+			usesDefaultSessionDir: false,
+		});
+
+		expect(formatResumeCommand(sessionManager)).toBe(
+			`${APP_NAME} --session-dir /tmp/custom-pi-sessions --session test-session`,
+		);
+	});
+
+	it("quotes session dirs containing spaces", () => {
+		setStdoutIsTTY(true);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({
+			sessionFile,
+			sessionId: "test-session",
+			sessionDir: "/tmp/custom pi sessions",
+			usesDefaultSessionDir: false,
+		});
+
+		expect(formatResumeCommand(sessionManager)).toBe(
+			`${APP_NAME} --session-dir '/tmp/custom pi sessions' --session test-session`,
+		);
+	});
+
+	it("quotes session dirs containing single quotes", () => {
+		setStdoutIsTTY(true);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({
+			sessionFile,
+			sessionId: "test-session",
+			sessionDir: "/tmp/custom pi's sessions",
+			usesDefaultSessionDir: false,
+		});
+
+		expect(formatResumeCommand(sessionManager)).toBe(
+			`${APP_NAME} --session-dir '/tmp/custom pi'\\''s sessions' --session test-session`,
+		);
+	});
+
+	it("returns undefined when stdout is not a TTY", () => {
+		setStdoutIsTTY(false);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({ sessionFile });
+
+		expect(formatResumeCommand(sessionManager)).toBeUndefined();
+	});
+
+	it("returns undefined for in-memory sessions", () => {
+		setStdoutIsTTY(true);
+		const sessionFile = createTempFile();
+		const sessionManager = createSessionManager({ persisted: false, sessionFile });
+
+		expect(formatResumeCommand(sessionManager)).toBeUndefined();
+	});
+
+	it("returns undefined when the session file is missing", () => {
+		setStdoutIsTTY(true);
+		const sessionManager = createSessionManager({ sessionFile: "/tmp/pi-missing-session.jsonl" });
+
+		expect(formatResumeCommand(sessionManager)).toBeUndefined();
+	});
+
+	it("returns undefined when the session file is not set", () => {
+		setStdoutIsTTY(true);
+		const sessionManager = createSessionManager({ sessionFile: undefined });
+
+		expect(formatResumeCommand(sessionManager)).toBeUndefined();
+	});
+});

packages/coding-agent/test/git-merge-and-resolve-extension.test.ts

@@ -0,0 +1,206 @@
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, it, vi } from "vitest";
+import mergeAndResolve from "../examples/extensions/git-merge-and-resolve.ts";
+import type { ExecResult, ExtensionAPI, ExtensionContext } from "../src/core/extensions/index.ts";
+
+type AgentEndHandler = (event: { type: "agent_end" }, ctx: ExtensionContext) => Promise<undefined>;
+
+const ok: ExecResult = { stdout: "", stderr: "", code: 0, killed: false };
+const fail: ExecResult = { stdout: "", stderr: "error", code: 1, killed: false };
+
+/** Standard exec results for a clean repo tracking origin/main, not in a merge. */
+function withUpstream(results: Map<string, ExecResult>): Map<string, ExecResult> {
+	results.set("git rev-parse --git-dir", ok);
+	results.set("git rev-parse MERGE_HEAD", fail);
+	results.set("git status --porcelain", ok);
+	results.set("git rev-parse --abbrev-ref --symbolic-full-name @{u}", { ...ok, stdout: "origin/main\n" });
+	results.set("git fetch origin", ok);
+	return results;
+}
+
+function setup(cwd: string, execResults: Map<string, ExecResult>) {
+	let handler: AgentEndHandler | undefined;
+	const sendUserMessage = vi.fn();
+
+	const exec = vi.fn<ExtensionAPI["exec"]>().mockImplementation(async (cmd, args) => {
+		const key = [cmd, ...args].join(" ");
+		return execResults.get(key) ?? fail;
+	});
+
+	const api = {
+		on: (event: string, h: AgentEndHandler) => {
+			if (event === "agent_end") handler = h;
+		},
+		exec,
+		sendUserMessage,
+	} as unknown as ExtensionAPI;
+
+	mergeAndResolve(api);
+
+	const ctx = { cwd, ui: { notify: vi.fn() } } as unknown as ExtensionContext;
+
+	async function trigger() {
+		await handler!({ type: "agent_end" }, ctx);
+	}
+
+	return { trigger, exec, sendUserMessage };
+}
+
+describe("git-merge-and-resolve example", () => {
+	let tempDir: string;
+
+	afterEach(() => {
+		if (tempDir) rmSync(tempDir, { recursive: true, force: true });
+	});
+
+	function createTempDir() {
+		tempDir = mkdtempSync(join(tmpdir(), "pi-merge-test-"));
+		return tempDir;
+	}
+
+	it("skips when not a git repository", async () => {
+		const cwd = createTempDir();
+		const results = new Map<string, ExecResult>();
+		results.set("git rev-parse --git-dir", fail);
+
+		const { trigger, exec, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(exec).toHaveBeenCalledTimes(1);
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+
+	it("skips when no upstream is configured", async () => {
+		const cwd = createTempDir();
+		const results = new Map<string, ExecResult>();
+		results.set("git rev-parse --git-dir", ok);
+		results.set("git rev-parse --abbrev-ref --symbolic-full-name @{u}", fail);
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+
+	it("re-sends conflicts when in an unfinished merge", async () => {
+		const cwd = createTempDir();
+		const conflictContent = ["<<<<<<< HEAD", "ours", "=======", "theirs", ">>>>>>> origin/main"].join("\n");
+		writeFileSync(join(cwd, "file.ts"), conflictContent);
+
+		const results = new Map<string, ExecResult>();
+		results.set("git rev-parse --git-dir", ok);
+		results.set("git rev-parse MERGE_HEAD", ok);
+		results.set("git diff --name-only --diff-filter=U", { ...ok, stdout: "file.ts\n" });
+
+		const { trigger, exec, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		// Should not attempt a new fetch/merge
+		expect(exec).not.toHaveBeenCalledWith("git", ["fetch", "origin"]);
+		expect(sendUserMessage).toHaveBeenCalledTimes(1);
+		const message = sendUserMessage.mock.calls[0][0] as string;
+		expect(message).toContain("file.ts:1-5");
+	});
+
+	it("skips when working tree is dirty and not in a merge", async () => {
+		const cwd = createTempDir();
+		const results = new Map<string, ExecResult>();
+		results.set("git rev-parse --git-dir", ok);
+		results.set("git rev-parse MERGE_HEAD", fail);
+		results.set("git status --porcelain", { ...ok, stdout: " M src/index.ts\n" });
+
+		const { trigger, exec, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(exec).not.toHaveBeenCalledWith("git", ["fetch", "origin"]);
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+
+	it("skips when fetch fails", async () => {
+		const cwd = createTempDir();
+		const results = withUpstream(new Map<string, ExecResult>());
+		results.set("git fetch origin", fail);
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+
+	it("skips when merge is clean", async () => {
+		const cwd = createTempDir();
+		const results = withUpstream(new Map<string, ExecResult>());
+		results.set("git merge --no-ff origin/main", ok);
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+
+	it("sends conflict report as a follow-up", async () => {
+		const cwd = createTempDir();
+		const conflictContent = [
+			"line 1",
+			"<<<<<<< HEAD",
+			"our change",
+			"=======",
+			"their change",
+			">>>>>>> origin/main",
+			"line 7",
+			"<<<<<<< HEAD",
+			"second conflict",
+			"=======",
+			"their second",
+			">>>>>>> origin/main",
+		].join("\n");
+
+		mkdirSync(join(cwd, "src"), { recursive: true });
+		writeFileSync(join(cwd, "src/index.ts"), conflictContent);
+
+		const results = withUpstream(new Map<string, ExecResult>());
+		results.set("git merge --no-ff origin/main", { ...fail, code: 1 });
+		results.set("git diff --name-only --diff-filter=U", { ...ok, stdout: "src/index.ts\n" });
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).toHaveBeenCalledTimes(1);
+		const [message, options] = sendUserMessage.mock.calls[0];
+		expect(message).toContain("src/index.ts:2-6 (ours 3, theirs 5)");
+		expect(message).toContain("src/index.ts:8-12 (ours 9, theirs 11)");
+		expect(options).toEqual({ deliverAs: "followUp" });
+	});
+
+	it("handles empty ours or theirs sections", async () => {
+		const cwd = createTempDir();
+		const conflictContent = ["<<<<<<< HEAD", "=======", "only theirs", ">>>>>>> origin/main"].join("\n");
+
+		writeFileSync(join(cwd, "empty-ours.ts"), conflictContent);
+
+		const results = withUpstream(new Map<string, ExecResult>());
+		results.set("git merge --no-ff origin/main", { ...fail, code: 1 });
+		results.set("git diff --name-only --diff-filter=U", { ...ok, stdout: "empty-ours.ts\n" });
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).toHaveBeenCalledTimes(1);
+		const message = sendUserMessage.mock.calls[0][0] as string;
+		expect(message).toContain("empty-ours.ts:1-4 (ours empty, theirs 3)");
+	});
+
+	it("skips message when merge fails but no conflict markers found", async () => {
+		const cwd = createTempDir();
+		const results = withUpstream(new Map<string, ExecResult>());
+		results.set("git merge --no-ff origin/main", { ...fail, code: 1 });
+		results.set("git diff --name-only --diff-filter=U", { ...ok, stdout: "" });
+
+		const { trigger, sendUserMessage } = setup(cwd, results);
+		await trigger();
+
+		expect(sendUserMessage).not.toHaveBeenCalled();
+	});
+});

packages/coding-agent/test/git-ssh-url.test.ts

@@ -62,6 +62,19 @@ describe("Git URL Parsing", () => {
 		});
 	});
 
+	it("should reject unsafe git install path inputs", () => {
+		for (const source of [
+			"git:git@evil.example:../../victim/repo",
+			"https://evil.example/..%2F..%2Fvictim/repo",
+			"https://evil.example/..%2F..%2Fvictim/repo%",
+			"git:git@evil.example:/absolute/repo",
+			"git:git@evil.example:user\\repo/name",
+			"git:git@evil.example:user/repo\0name",
+		]) {
+			expect(parseGitUrl(source)).toBeNull();
+		}
+	});
+
 	describe("unsupported without git: prefix", () => {
 		it("should reject git@host:path without git: prefix", () => {
 			expect(parseGitUrl("git@github.com:user/repo")).toBeNull();

packages/coding-agent/test/git-update.test.ts

@@ -7,7 +7,6 @@
  */
 
 import { spawnSync } from "node:child_process";
-import { createHash } from "node:crypto";
 import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
@@ -51,6 +50,20 @@ function getFileContent(repoDir: string, filename: string): string {
 	return readFileSync(join(repoDir, filename), "utf-8");
 }
 
+type GitSourceForTest = {
+	type: "git";
+	repo: string;
+	host: string;
+	path: string;
+	pinned: boolean;
+	ref?: string;
+};
+
+interface PackageManagerPathInternals {
+	parseSource(source: string): GitSourceForTest;
+	getGitInstallPath(source: GitSourceForTest, scope: "temporary"): string;
+}
+
 describe("DefaultPackageManager git update", () => {
 	let tempDir: string;
 	let remoteDir: string; // Simulates the "remote" repository
@@ -282,7 +295,7 @@ describe("DefaultPackageManager git update", () => {
 	});
 
 	describe("pinned sources", () => {
-		it("should not update pinned git sources (with @ref)", async () => {
+		it("should not move pinned git sources past their configured ref", async () => {
 			// Create remote repo first to get the initial commit
 			mkdirSync(remoteDir, { recursive: true });
 			initGitRepo(remoteDir);
@@ -301,21 +314,83 @@ describe("DefaultPackageManager git update", () => {
 			// Add new commit to remote
 			createCommit(remoteDir, "extension.ts", "// v2", "Second commit");
 
-			// Update should be skipped for pinned sources
 			await packageManager.update();
 
 			// Should still be on initial commit
 			expect(getCurrentCommit(installedDir)).toBe(initialCommit);
 			expect(getFileContent(installedDir, "extension.ts")).toBe("// v1");
 		});
+
+		it("should checkout the configured pinned git ref during full and targeted updates", async () => {
+			mkdirSync(remoteDir, { recursive: true });
+			initGitRepo(remoteDir);
+			const v1Commit = createCommit(remoteDir, "extension.ts", "// v1", "Initial commit");
+			git(["tag", "v1"], remoteDir);
+			const v2Commit = createCommit(remoteDir, "extension.ts", "// v2", "Second commit");
+			git(["tag", "v2"], remoteDir);
+
+			mkdirSync(join(agentDir, "git", "github.com", "test"), { recursive: true });
+			git(["clone", remoteDir, installedDir], tempDir);
+			git(["checkout", "v1"], installedDir);
+			expect(getCurrentCommit(installedDir)).toBe(v1Commit);
+
+			const pinnedSource = `${gitSource}@v2`;
+			settingsManager.setPackages([pinnedSource]);
+
+			await packageManager.update();
+
+			expect(getCurrentCommit(installedDir)).toBe(v2Commit);
+			expect(getFileContent(installedDir, "extension.ts")).toBe("// v2");
+
+			git(["checkout", "v1"], installedDir);
+
+			await packageManager.update(pinnedSource);
+
+			expect(getCurrentCommit(installedDir)).toBe(v2Commit);
+			expect(getFileContent(installedDir, "extension.ts")).toBe("// v2");
+		});
+
+		it("should not reset an annotated tag checkout that already matches the configured ref", async () => {
+			mkdirSync(remoteDir, { recursive: true });
+			initGitRepo(remoteDir);
+			const taggedCommit = createCommit(remoteDir, "extension.ts", "// v1", "Initial commit");
+			git(["tag", "-a", "v1", "-m", "v1"], remoteDir);
+
+			mkdirSync(join(agentDir, "git", "github.com", "test"), { recursive: true });
+			git(["clone", remoteDir, installedDir], tempDir);
+			git(["checkout", "v1"], installedDir);
+			expect(getCurrentCommit(installedDir)).toBe(taggedCommit);
+
+			settingsManager.setPackages([`${gitSource}@v1`]);
+
+			const executedCommands: string[] = [];
+			const managerWithInternals = packageManager as unknown as {
+				runCommand: (command: string, args: string[], options?: { cwd?: string }) => Promise<void>;
+			};
+			managerWithInternals.runCommand = async (command, args, options) => {
+				executedCommands.push(`${command} ${args.join(" ")}`);
+				const result = spawnSync(command, args, {
+					cwd: options?.cwd,
+					encoding: "utf-8",
+				});
+				if (result.status !== 0) {
+					throw new Error(`Command failed: ${command} ${args.join(" ")}\n${result.stderr}`);
+				}
+			};
+
+			await packageManager.update();
+
+			expect(executedCommands).toContain("git fetch origin v1");
+			expect(executedCommands.some((command) => command.startsWith("git reset --hard"))).toBe(false);
+			expect(executedCommands).not.toContain("git clean -fdx");
+			expect(getCurrentCommit(installedDir)).toBe(taggedCommit);
+		});
 	});
 
 	describe("temporary git sources", () => {
 		it("should refresh cached temporary git sources when resolving", async () => {
-			const gitHost = "github.com";
-			const gitPath = "test/extension";
-			const hash = createHash("sha256").update(`git-${gitHost}-${gitPath}`).digest("hex").slice(0, 8);
-			const cachedDir = join(tmpdir(), "pi-extensions", `git-${gitHost}`, hash, gitPath);
+			const managerWithPaths = packageManager as unknown as PackageManagerPathInternals;
+			const cachedDir = managerWithPaths.getGitInstallPath(managerWithPaths.parseSource(gitSource), "temporary");
 			const extensionFile = join(cachedDir, "pi-extensions", "session-breakdown.ts");
 
 			rmSync(cachedDir, { recursive: true, force: true });
@@ -359,10 +434,8 @@ describe("DefaultPackageManager git update", () => {
 		});
 
 		it("should not refresh pinned temporary git sources", async () => {
-			const gitHost = "github.com";
-			const gitPath = "test/extension";
-			const hash = createHash("sha256").update(`git-${gitHost}-${gitPath}`).digest("hex").slice(0, 8);
-			const cachedDir = join(tmpdir(), "pi-extensions", `git-${gitHost}`, hash, gitPath);
+			const managerWithPaths = packageManager as unknown as PackageManagerPathInternals;
+			const cachedDir = managerWithPaths.getGitInstallPath(managerWithPaths.parseSource(gitSource), "temporary");
 			const extensionFile = join(cachedDir, "pi-extensions", "session-breakdown.ts");
 
 			rmSync(cachedDir, { recursive: true, force: true });

packages/coding-agent/test/http-dispatcher.test.ts

@@ -0,0 +1,53 @@
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { applyHttpProxySettings } from "../src/core/http-dispatcher.ts";
+
+const PROXY_ENV_KEYS = ["HTTP_PROXY", "HTTPS_PROXY"] as const;
+
+describe("http proxy settings", () => {
+	let savedEnv: Record<(typeof PROXY_ENV_KEYS)[number], string | undefined>;
+
+	beforeEach(() => {
+		savedEnv = Object.fromEntries(PROXY_ENV_KEYS.map((key) => [key, process.env[key]])) as Record<
+			(typeof PROXY_ENV_KEYS)[number],
+			string | undefined
+		>;
+		for (const key of PROXY_ENV_KEYS) {
+			delete process.env[key];
+		}
+	});
+
+	afterEach(() => {
+		for (const key of PROXY_ENV_KEYS) {
+			const value = savedEnv[key];
+			if (value === undefined) {
+				delete process.env[key];
+			} else {
+				process.env[key] = value;
+			}
+		}
+	});
+
+	it("applies httpProxy to HTTP_PROXY and HTTPS_PROXY", () => {
+		applyHttpProxySettings("http://127.0.0.1:7890");
+
+		expect(process.env.HTTP_PROXY).toBe("http://127.0.0.1:7890");
+		expect(process.env.HTTPS_PROXY).toBe("http://127.0.0.1:7890");
+	});
+
+	it("does not override existing proxy env vars", () => {
+		process.env.HTTP_PROXY = "http://env-http:8080";
+		process.env.HTTPS_PROXY = "http://env-https:8080";
+
+		applyHttpProxySettings("http://settings:7890");
+
+		expect(process.env.HTTP_PROXY).toBe("http://env-http:8080");
+		expect(process.env.HTTPS_PROXY).toBe("http://env-https:8080");
+	});
+
+	it("ignores empty values", () => {
+		applyHttpProxySettings("   ");
+
+		expect(process.env.HTTP_PROXY).toBeUndefined();
+		expect(process.env.HTTPS_PROXY).toBeUndefined();
+	});
+});

packages/coding-agent/test/image-processing.test.ts

@@ -22,6 +22,10 @@ const MEDIUM_PNG_100x100 =
 const LARGE_PNG_200x200 =
 	"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";
 
+function imageBytes(base64Data: string): Uint8Array {
+	return Buffer.from(base64Data, "base64");
+}
+
 describe("convertToPng", () => {
 	it("should return original data for PNG input", async () => {
 		const result = await convertToPng(TINY_PNG, "image/png");
@@ -46,11 +50,28 @@ describe("convertToPng", () => {
 });
 
 describe("resizeImage", () => {
+	it("should keep caller input bytes intact", async () => {
+		const input = new Uint8Array(imageBytes(TINY_PNG));
+		const originalByteLength = input.byteLength;
+		const originalFirstByte = input[0];
+
+		const result = await resizeImage(input, "image/png", {
+			maxWidth: 100,
+			maxHeight: 100,
+			maxBytes: 1024 * 1024,
+		});
+
+		expect(result).not.toBeNull();
+		expect(input.byteLength).toBe(originalByteLength);
+		expect(input[0]).toBe(originalFirstByte);
+	});
+
 	it("should return original image if within limits", async () => {
-		const result = await resizeImage(
-			{ type: "image", data: TINY_PNG, mimeType: "image/png" },
-			{ maxWidth: 100, maxHeight: 100, maxBytes: 1024 * 1024 },
-		);
+		const result = await resizeImage(imageBytes(TINY_PNG), "image/png", {
+			maxWidth: 100,
+			maxHeight: 100,
+			maxBytes: 1024 * 1024,
+		});
 
 		expect(result).not.toBeNull();
 		expect(result!.wasResized).toBe(false);
@@ -62,10 +83,11 @@ describe("resizeImage", () => {
 	});
 
 	it("should resize image exceeding dimension limits", async () => {
-		const result = await resizeImage(
-			{ type: "image", data: MEDIUM_PNG_100x100, mimeType: "image/png" },
-			{ maxWidth: 50, maxHeight: 50, maxBytes: 1024 * 1024 },
-		);
+		const result = await resizeImage(imageBytes(MEDIUM_PNG_100x100), "image/png", {
+			maxWidth: 50,
+			maxHeight: 50,
+			maxBytes: 1024 * 1024,
+		});
 
 		expect(result).not.toBeNull();
 		expect(result!.wasResized).toBe(true);
@@ -80,10 +102,11 @@ describe("resizeImage", () => {
 		const originalSize = originalBuffer.length;
 
 		// Set maxBytes to less than the original encoded image size
-		const result = await resizeImage(
-			{ type: "image", data: LARGE_PNG_200x200, mimeType: "image/png" },
-			{ maxWidth: 2000, maxHeight: 2000, maxBytes: Math.floor(LARGE_PNG_200x200.length * 0.9) },
-		);
+		const result = await resizeImage(imageBytes(LARGE_PNG_200x200), "image/png", {
+			maxWidth: 2000,
+			maxHeight: 2000,
+			maxBytes: Math.floor(LARGE_PNG_200x200.length * 0.9),
+		});
 
 		// Should have tried to reduce size
 		expect(result).not.toBeNull();
@@ -93,19 +116,21 @@ describe("resizeImage", () => {
 	});
 
 	it("should return null when image cannot be resized below maxBytes", async () => {
-		const result = await resizeImage(
-			{ type: "image", data: LARGE_PNG_200x200, mimeType: "image/png" },
-			{ maxWidth: 2000, maxHeight: 2000, maxBytes: 1 },
-		);
+		const result = await resizeImage(imageBytes(LARGE_PNG_200x200), "image/png", {
+			maxWidth: 2000,
+			maxHeight: 2000,
+			maxBytes: 1,
+		});
 
 		expect(result).toBeNull();
 	});
 
 	it("should handle JPEG input", async () => {
-		const result = await resizeImage(
-			{ type: "image", data: TINY_JPEG, mimeType: "image/jpeg" },
-			{ maxWidth: 100, maxHeight: 100, maxBytes: 1024 * 1024 },
-		);
+		const result = await resizeImage(imageBytes(TINY_JPEG), "image/jpeg", {
+			maxWidth: 100,
+			maxHeight: 100,
+			maxBytes: 1024 * 1024,
+		});
 
 		expect(result).not.toBeNull();
 		expect(result!.wasResized).toBe(false);

packages/coding-agent/test/input-transform-streaming-example.test.ts

@@ -0,0 +1,84 @@
+import { describe, expect, it, vi } from "vitest";
+import inputTransformStreaming from "../examples/extensions/input-transform-streaming.ts";
+import type {
+	ExecResult,
+	ExtensionAPI,
+	ExtensionContext,
+	InputEvent,
+	InputEventResult,
+} from "../src/core/extensions/index.ts";
+
+type InputHandler = (event: InputEvent, ctx: ExtensionContext) => Promise<InputEventResult | undefined>;
+
+function setup(execResult: ExecResult) {
+	let handler: InputHandler | undefined;
+
+	const exec = vi.fn<ExtensionAPI["exec"]>().mockResolvedValue(execResult);
+
+	const api = {
+		on: (event: string, h: InputHandler) => {
+			if (event === "input") handler = h;
+		},
+		exec,
+	} as unknown as ExtensionAPI;
+
+	inputTransformStreaming(api);
+
+	const ctx = {} as ExtensionContext;
+
+	function emit(text: string, streamingBehavior?: "steer" | "followUp") {
+		return handler!({ type: "input", text, source: "interactive", streamingBehavior }, ctx);
+	}
+
+	return { emit, exec };
+}
+
+describe("input-transform-streaming example", () => {
+	const diffOutput = " src/index.ts | 5 ++---\n 1 file changed, 2 insertions(+), 3 deletions(-)";
+	const gitSuccess: ExecResult = { stdout: diffOutput, stderr: "", code: 0, killed: false };
+	const gitEmpty: ExecResult = { stdout: "", stderr: "", code: 0, killed: false };
+	const gitFail: ExecResult = { stdout: "", stderr: "not a git repo", code: 128, killed: false };
+
+	it("skips exec during steering", async () => {
+		const { emit, exec } = setup(gitSuccess);
+		const result = await emit("what changes did I make?", "steer");
+		expect(result).toEqual({ action: "continue" });
+		expect(exec).not.toHaveBeenCalled();
+	});
+
+	it("transforms when idle and text matches trigger", async () => {
+		const { emit, exec } = setup(gitSuccess);
+		const result = await emit("review my changes");
+		expect(exec).toHaveBeenCalledWith("git", ["diff", "--stat"]);
+		expect(result).toMatchObject({ action: "transform" });
+		const text = (result as { text: string }).text;
+		expect(text).toContain("review my changes");
+		expect(text).toContain("src/index.ts");
+	});
+
+	it("transforms when queued as follow-up", async () => {
+		const { emit, exec } = setup(gitSuccess);
+		const result = await emit("show me the diff", "followUp");
+		expect(exec).toHaveBeenCalled();
+		expect(result).toMatchObject({ action: "transform" });
+	});
+
+	it("continues when text does not match trigger", async () => {
+		const { emit, exec } = setup(gitSuccess);
+		const result = await emit("explain this function");
+		expect(result).toEqual({ action: "continue" });
+		expect(exec).not.toHaveBeenCalled();
+	});
+
+	it("continues when git diff is empty", async () => {
+		const { emit } = setup(gitEmpty);
+		const result = await emit("any changes?");
+		expect(result).toEqual({ action: "continue" });
+	});
+
+	it("continues when git fails", async () => {
+		const { emit } = setup(gitFail);
+		const result = await emit("show modified files");
+		expect(result).toEqual({ action: "continue" });
+	});
+});

packages/coding-agent/test/interactive-mode-startup-input.test.ts

@@ -0,0 +1,72 @@
+import { describe, expect, it, vi } from "vitest";
+import { InteractiveMode } from "../src/modes/interactive/interactive-mode.ts";
+
+type SubmitContext = {
+	defaultEditor: { onSubmit?: (text: string) => void };
+	editor: {
+		addToHistory?: (text: string) => void;
+		setText: (text: string) => void;
+	};
+	session: {
+		isCompacting: boolean;
+		isStreaming: boolean;
+		isBashRunning: boolean;
+		prompt: (text: string, options?: unknown) => Promise<void>;
+	};
+	flushPendingBashComponents: () => void;
+	onInputCallback?: (text: string) => void;
+	pendingUserInputs: string[];
+};
+
+type InputContext = {
+	onInputCallback?: (text: string) => void;
+	pendingUserInputs: string[];
+};
+
+type InteractiveModePrivate = {
+	setupEditorSubmitHandler(this: SubmitContext): void;
+	getUserInput(this: InputContext): Promise<string>;
+};
+
+const interactiveModePrototype = InteractiveMode.prototype as unknown as InteractiveModePrivate;
+
+function createSubmitContext(): SubmitContext {
+	return {
+		defaultEditor: {},
+		editor: {
+			addToHistory: vi.fn(),
+			setText: vi.fn(),
+		},
+		session: {
+			isCompacting: false,
+			isStreaming: false,
+			isBashRunning: false,
+			prompt: vi.fn(async () => {}),
+		},
+		flushPendingBashComponents: vi.fn(),
+		pendingUserInputs: [],
+	};
+}
+
+describe("InteractiveMode startup input", () => {
+	it("queues a normal prompt submitted before the input callback is installed", async () => {
+		const context = createSubmitContext();
+		interactiveModePrototype.setupEditorSubmitHandler.call(context);
+
+		await context.defaultEditor.onSubmit?.(" early prompt ");
+
+		expect(context.pendingUserInputs).toEqual(["early prompt"]);
+		expect(context.flushPendingBashComponents).toHaveBeenCalledTimes(1);
+		expect(context.editor.addToHistory).toHaveBeenCalledWith("early prompt");
+	});
+
+	it("returns queued startup input before installing a new input callback", async () => {
+		const context: InputContext = {
+			pendingUserInputs: ["queued prompt"],
+		};
+
+		await expect(interactiveModePrototype.getUserInput.call(context)).resolves.toBe("queued prompt");
+		expect(context.onInputCallback).toBeUndefined();
+		expect(context.pendingUserInputs).toEqual([]);
+	});
+});

packages/coding-agent/test/interactive-mode-status.test.ts

@@ -1,7 +1,9 @@
 import { homedir } from "node:os";
 import * as path from "node:path";
-import { type AutocompleteProvider, CombinedAutocompleteProvider, Container } from "@earendil-works/pi-tui";
+import { type AutocompleteProvider, CombinedAutocompleteProvider } from "@earendil-works/pi-tui";
 import { beforeAll, describe, expect, test, vi } from "vitest";
+import { type Component, Container, type Focusable, TUI } from "../../tui/src/tui.ts";
+import { VirtualTerminal } from "../../tui/test/virtual-terminal.ts";
 import type { AutocompleteProviderFactory } from "../src/core/extensions/types.ts";
 import type { SourceInfo } from "../src/core/source-info.ts";
 import { InteractiveMode } from "../src/modes/interactive/interactive-mode.ts";
@@ -17,6 +19,41 @@ function renderAll(container: Container, width = 120): string {
 	return container.children.flatMap((child) => child.render(width)).join("\n");
 }
 
+class TestFocusableComponent implements Component, Focusable {
+	focused = false;
+	inputs: string[] = [];
+	private readonly label: string;
+	private text = "";
+
+	constructor(label: string) {
+		this.label = label;
+	}
+
+	handleInput(data: string): void {
+		this.inputs.push(data);
+	}
+
+	getText(): string {
+		return this.text;
+	}
+
+	setText(text: string): void {
+		this.text = text;
+	}
+
+	render(): string[] {
+		return [this.label];
+	}
+
+	invalidate(): void {}
+}
+
+async function flushTui(tui: TUI, terminal: VirtualTerminal): Promise<void> {
+	tui.requestRender(true);
+	await Promise.resolve();
+	await terminal.waitForRender();
+}
+
 function normalizeRenderedOutput(container: Container, width = 220): string {
 	return renderAll(container, width)
 		.replace(/\u001b\[[0-9;]*m/g, "")
@@ -114,6 +151,13 @@ describe("InteractiveMode.createExtensionUIContext setTheme", () => {
 		const fakeThis: any = {
 			session: { settingsManager },
 			settingsManager,
+			themeController: {
+				setThemeInstance: vi.fn(() => ({ success: true })),
+				setThemeName: vi.fn(() => {
+					fakeThis.ui.requestRender();
+					return { success: true };
+				}),
+			},
 			ui: { requestRender: vi.fn() },
 		};
 
@@ -121,6 +165,7 @@ describe("InteractiveMode.createExtensionUIContext setTheme", () => {
 		const result = uiContext.setTheme("light");
 
 		expect(result.success).toBe(true);
+		expect(fakeThis.themeController.setThemeName).toHaveBeenCalledWith("light");
 		expect(settingsManager.setTheme).toHaveBeenCalledWith("light");
 		expect(currentTheme).toBe("light");
 		expect(fakeThis.ui.requestRender).toHaveBeenCalledTimes(1);
@@ -136,6 +181,10 @@ describe("InteractiveMode.createExtensionUIContext setTheme", () => {
 		const fakeThis: any = {
 			session: { settingsManager },
 			settingsManager,
+			themeController: {
+				setThemeInstance: vi.fn(() => ({ success: true })),
+				setThemeName: vi.fn(() => ({ success: false, error: "Theme not found" })),
+			},
 			ui: { requestRender: vi.fn() },
 		};
 
@@ -143,11 +192,84 @@ describe("InteractiveMode.createExtensionUIContext setTheme", () => {
 		const result = uiContext.setTheme("__missing_theme__");
 
 		expect(result.success).toBe(false);
+		expect(fakeThis.themeController.setThemeName).toHaveBeenCalledWith("__missing_theme__");
 		expect(settingsManager.setTheme).not.toHaveBeenCalled();
 		expect(fakeThis.ui.requestRender).not.toHaveBeenCalled();
 	});
 });
 
+describe("InteractiveMode.showExtensionCustom", () => {
+	beforeAll(() => {
+		initTheme("dark");
+	});
+
+	test("overlay custom UI reclaims input after non-overlay custom UI closes", async () => {
+		const terminal = new VirtualTerminal(80, 24);
+		const ui = new TUI(terminal);
+		const editorContainer = new Container();
+		const editor = new TestFocusableComponent("EDITOR");
+		const palette = new TestFocusableComponent("PALETTE");
+		const overlay = new TestFocusableComponent("OVERLAY");
+		const replacement = new TestFocusableComponent("REPLACEMENT");
+		let closeOverlay: (value: string) => void = () => {
+			throw new Error("closeOverlay was not initialized");
+		};
+		let closeReplacement: (value: string) => void = () => {
+			throw new Error("closeReplacement was not initialized");
+		};
+		const fakeThis = {
+			editor,
+			editorContainer,
+			keybindings: {},
+			ui,
+		};
+		const showExtensionCustom = <T>(
+			factory: (tui: TUI, theme: unknown, keybindings: unknown, done: (result: T) => void) => Component,
+			options?: { overlay?: boolean },
+		): Promise<T> =>
+			(InteractiveMode as any).prototype.showExtensionCustom.call(fakeThis, factory, options) as Promise<T>;
+
+		editorContainer.addChild(editor);
+		ui.addChild(editorContainer);
+		ui.addChild(palette);
+		ui.setFocus(palette);
+		ui.start();
+		try {
+			const overlayPromise = showExtensionCustom<string>(
+				(_tui, _theme, _keybindings, done) => {
+					closeOverlay = done;
+					return overlay;
+				},
+				{ overlay: true },
+			);
+			await flushTui(ui, terminal);
+			expect(overlay.focused).toBe(true);
+
+			const replacementPromise = showExtensionCustom<string>((_tui, _theme, _keybindings, done) => {
+				closeReplacement = done;
+				return replacement;
+			});
+			await flushTui(ui, terminal);
+			expect(replacement.focused).toBe(true);
+
+			closeReplacement("done");
+			await replacementPromise;
+			await flushTui(ui, terminal);
+			terminal.sendInput("x");
+			await flushTui(ui, terminal);
+
+			expect(overlay.inputs).toEqual(["x"]);
+			expect(editor.inputs).toEqual([]);
+			expect(overlay.focused).toBe(true);
+
+			closeOverlay("closed");
+			await overlayPromise;
+		} finally {
+			ui.stop();
+		}
+	});
+});
+
 describe("InteractiveMode.createExtensionUIContext addAutocompleteProvider", () => {
 	test("stores wrapper factories and rebuilds autocomplete immediately", () => {
 		const wrapper: AutocompleteProviderFactory = (current) => current;
@@ -215,6 +337,89 @@ describe("InteractiveMode.setupAutocompleteProvider", () => {
 		expect(provider.shouldTriggerFileCompletion?.(["foo"], 0, 3)).toBe(true);
 		expect(calls).toEqual(["shouldTrigger:wrap2", "shouldTrigger:wrap1"]);
 	});
+
+	test("merges triggerCharacters from wrapper factories", () => {
+		const defaultEditor = { setAutocompleteProvider: vi.fn() };
+		const customEditor = { setAutocompleteProvider: vi.fn() };
+		const passThrough =
+			(triggerCharacters: string[]): AutocompleteProviderFactory =>
+			(current) => ({
+				triggerCharacters,
+				getSuggestions: (lines, cursorLine, cursorCol, options) =>
+					current.getSuggestions(lines, cursorLine, cursorCol, options),
+				applyCompletion: (lines, cursorLine, cursorCol, item, prefix) =>
+					current.applyCompletion(lines, cursorLine, cursorCol, item, prefix),
+			});
+
+		const fakeThis = {
+			createBaseAutocompleteProvider: () => new CombinedAutocompleteProvider([], "/tmp/project", undefined),
+			defaultEditor,
+			editor: customEditor,
+			autocompleteProviderWrappers: [passThrough(["$"]), passThrough(["!"])],
+		};
+
+		(
+			InteractiveMode as unknown as {
+				prototype: { setupAutocompleteProvider: (this: typeof fakeThis) => void };
+			}
+		).prototype.setupAutocompleteProvider.call(fakeThis);
+
+		const provider = defaultEditor.setAutocompleteProvider.mock.calls[0]?.[0] as AutocompleteProvider;
+		expect(provider.triggerCharacters).toEqual(["$", "!"]);
+	});
+});
+
+describe("InteractiveMode.createBaseAutocompleteProvider", () => {
+	test("matches model command arguments across provider/model order", async () => {
+		type TestModel = { id: string; provider: string; name: string };
+		type FakeInteractiveMode = {
+			session: {
+				scopedModels: Array<{ model: TestModel }>;
+				modelRegistry: { getAvailable: () => TestModel[] };
+				promptTemplates: [];
+				extensionRunner: { getRegisteredCommands: () => [] };
+				resourceLoader: { getSkills: () => { skills: [] } };
+			};
+			settingsManager: { getEnableSkillCommands: () => boolean };
+			skillCommands: Map<string, string>;
+			sessionManager: { getCwd: () => string };
+			fdPath: null;
+		};
+
+		const createBaseAutocompleteProvider = (
+			InteractiveMode as unknown as {
+				prototype: { createBaseAutocompleteProvider(this: FakeInteractiveMode): AutocompleteProvider };
+			}
+		).prototype.createBaseAutocompleteProvider;
+		const models = [
+			{ id: "gpt-5.2-codex", provider: "github-copilot", name: "GPT-5.2 Codex" },
+			{ id: "gpt-5.5", provider: "openai-codex", name: "GPT-5.5" },
+		];
+		const fakeThis: FakeInteractiveMode = {
+			session: {
+				scopedModels: [],
+				modelRegistry: { getAvailable: () => models },
+				promptTemplates: [],
+				extensionRunner: { getRegisteredCommands: () => [] },
+				resourceLoader: { getSkills: () => ({ skills: [] }) },
+			},
+			settingsManager: { getEnableSkillCommands: () => false },
+			skillCommands: new Map(),
+			sessionManager: { getCwd: () => "/tmp" },
+			fdPath: null,
+		};
+
+		const provider = createBaseAutocompleteProvider.call(fakeThis);
+		const line = "/model codexgpt";
+		const suggestions = await provider.getSuggestions([line], 0, line.length, {
+			signal: new AbortController().signal,
+		});
+
+		expect(suggestions?.items.map((item) => item.value)).toEqual([
+			"openai-codex/gpt-5.5",
+			"github-copilot/gpt-5.2-codex",
+		]);
+	});
 });
 
 describe("InteractiveMode.showLoadedResources", () => {

packages/coding-agent/test/model-registry.test.ts

@@ -4,7 +4,7 @@ import { join } from "node:path";
 import type { AnthropicMessagesCompat, Api, Context, Model, OpenAICompletionsCompat } from "@earendil-works/pi-ai";
 import { getApiProvider } from "@earendil-works/pi-ai";
 import { getOAuthProvider } from "@earendil-works/pi-ai/oauth";
-import { afterEach, beforeEach, describe, expect, test } from "vitest";
+import { afterEach, beforeEach, describe, expect, test, vi } from "vitest";
 import { AuthStorage } from "../src/core/auth-storage.ts";
 import { clearApiKeyCache, ModelRegistry, type ProviderConfigInput } from "../src/core/model-registry.ts";
 
@@ -25,6 +25,7 @@ describe("ModelRegistry", () => {
 			rmSync(tempDir, { recursive: true });
 		}
 		clearApiKeyCache();
+		vi.restoreAllMocks();
 	});
 
 	/** Create minimal provider config  */
@@ -35,7 +36,7 @@ describe("ModelRegistry", () => {
 	): ProviderConfigInput {
 		return {
 			baseUrl,
-			apiKey: "TEST_KEY",
+			apiKey: "test-key",
 			api: api as Api,
 			models: models.map((m) => ({
 				id: m.id,
@@ -433,6 +434,43 @@ describe("ModelRegistry", () => {
 			expect(compat?.cacheControlFormat).toBe("anthropic");
 		});
 
+		test("compat schema accepts chat template thinking configuration", () => {
+			writeRawModelsJson({
+				demo: {
+					baseUrl: "https://example.com/v1",
+					apiKey: "DEMO_KEY",
+					api: "openai-completions",
+					models: [
+						{
+							id: "demo-model",
+							reasoning: true,
+							input: ["text"],
+							cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+							contextWindow: 1000,
+							maxTokens: 100,
+							compat: {
+								thinkingFormat: "chat-template",
+								chatTemplateKwargs: {
+									preserve_thinking: true,
+									thinking: { $var: "thinking.enabled" },
+								},
+							},
+						},
+					],
+				},
+			});
+
+			const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+			const compat = registry.find("demo", "demo-model")?.compat as OpenAICompletionsCompat | undefined;
+
+			expect(registry.getError()).toBeUndefined();
+			expect(compat?.thinkingFormat).toBe("chat-template");
+			expect(compat?.chatTemplateKwargs).toEqual({
+				preserve_thinking: true,
+				thinking: { $var: "thinking.enabled" },
+			});
+		});
+
 		test("compat schema accepts Anthropic eager tool input streaming flag", () => {
 			writeRawModelsJson({
 				demo: {
@@ -852,7 +890,7 @@ describe("ModelRegistry", () => {
 			registry.registerProvider("named-provider", {
 				name: "Named Provider",
 				baseUrl: "https://provider.test/v1",
-				apiKey: "TEST_KEY",
+				apiKey: "test-key",
 				api: "openai-completions",
 				models: [
 					{
@@ -892,6 +930,91 @@ describe("ModelRegistry", () => {
 			expect(registry.getProviderDisplayName("oauth-provider")).toBe("OAuth Provider");
 		});
 
+		test("stored API key env propagates to request auth and resolves headers", async () => {
+			authStorage.set("cloudflare-ai-gateway", {
+				type: "api_key",
+				key: "$CLOUDFLARE_API_KEY",
+				env: {
+					CLOUDFLARE_API_KEY: "stored-cf-token",
+					CLOUDFLARE_ACCOUNT_ID: "stored-account",
+				},
+			});
+			writeRawModelsJson({
+				"cloudflare-ai-gateway": {
+					headers: { "x-account": "$CLOUDFLARE_ACCOUNT_ID" },
+				},
+			});
+
+			const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+			const model = registry.getAll().find((m) => m.provider === "cloudflare-ai-gateway");
+			expect(model).toBeDefined();
+
+			const auth = await registry.getApiKeyAndHeaders(model!);
+
+			expect(auth).toEqual({
+				ok: true,
+				apiKey: "stored-cf-token",
+				headers: { "x-account": "stored-account" },
+				env: {
+					CLOUDFLARE_API_KEY: "stored-cf-token",
+					CLOUDFLARE_ACCOUNT_ID: "stored-account",
+				},
+			});
+		});
+
+		test("registerProvider treats uppercase apiKey and headers as literals", async () => {
+			const envKeys = ["CUSTOM_NAME", "BEARER", "MODEL_TOKEN"];
+			const savedEnv: Record<string, string | undefined> = {};
+			for (const key of envKeys) {
+				savedEnv[key] = process.env[key];
+				process.env[key] = `env-${key}`;
+			}
+			const warnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+
+			try {
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+
+				registry.registerProvider("literal-provider", {
+					...providerConfig("https://provider.test/v1", [{ id: "demo-model" }], "openai-completions"),
+					apiKey: "CUSTOM_NAME",
+					headers: { Authorization: "BEARER" },
+					models: [
+						{
+							id: "demo-model",
+							name: "demo-model",
+							reasoning: false,
+							input: ["text"],
+							cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+							contextWindow: 100000,
+							maxTokens: 8000,
+							headers: { "x-model-token": "MODEL_TOKEN" },
+						},
+					],
+				});
+
+				expect(await registry.getApiKeyForProvider("literal-provider")).toBe("CUSTOM_NAME");
+				const model = registry.find("literal-provider", "demo-model");
+				expect(model).toBeDefined();
+				expect(await registry.getApiKeyAndHeaders(model!)).toMatchObject({
+					ok: true,
+					apiKey: "CUSTOM_NAME",
+					headers: {
+						Authorization: "BEARER",
+						"x-model-token": "MODEL_TOKEN",
+					},
+				});
+				expect(warnSpy).not.toHaveBeenCalled();
+			} finally {
+				for (const key of envKeys) {
+					if (savedEnv[key] === undefined) {
+						delete process.env[key];
+					} else {
+						process.env[key] = savedEnv[key];
+					}
+				}
+			}
+		});
+
 		test("failed registerProvider does not persist invalid streamSimple config", () => {
 			const registry = ModelRegistry.create(authStorage, modelsJsonPath);
 
@@ -911,7 +1034,7 @@ describe("ModelRegistry", () => {
 
 			registry.registerProvider("demo-provider", {
 				baseUrl: "https://provider.test/v1",
-				apiKey: "TEST_KEY",
+				apiKey: "test-key",
 				api: "openai-completions",
 				models: [
 					{
@@ -931,7 +1054,7 @@ describe("ModelRegistry", () => {
 			expect(() =>
 				registry.registerProvider("demo-provider", {
 					baseUrl: "https://provider.test/v2",
-					apiKey: "TEST_KEY",
+					apiKey: "test-key",
 					models: [
 						{
 							id: "broken-model",
@@ -1187,7 +1310,117 @@ describe("ModelRegistry", () => {
 			expect(apiKey).toBeUndefined();
 		});
 
-		test("apiKey as environment variable name resolves to env value", async () => {
+		test("apiKey with $ prefix resolves to env value", async () => {
+			const originalEnv = process.env.TEST_API_KEY_12345;
+			process.env.TEST_API_KEY_12345 = "env-api-key-value";
+
+			try {
+				writeRawModelsJson({
+					"custom-provider": providerWithApiKey("$TEST_API_KEY_12345"),
+				});
+
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+				const apiKey = await registry.getApiKeyForProvider("custom-provider");
+
+				expect(apiKey).toBe("env-api-key-value");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_API_KEY_12345;
+				} else {
+					process.env.TEST_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("apiKey with braced env syntax resolves to env value", async () => {
+			const originalEnv = process.env.TEST_BRACED_API_KEY_12345;
+			process.env.TEST_BRACED_API_KEY_12345 = "braced-env-api-key-value";
+			const bracedKey = "$" + "{TEST_BRACED_API_KEY_12345}";
+
+			try {
+				writeRawModelsJson({
+					"custom-provider": providerWithApiKey(bracedKey),
+				});
+
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+				const apiKey = await registry.getApiKeyForProvider("custom-provider");
+
+				expect(apiKey).toBe("braced-env-api-key-value");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_BRACED_API_KEY_12345;
+				} else {
+					process.env.TEST_BRACED_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("apiKey interpolates braced env references inside literals", async () => {
+			const originalPartA = process.env.TEST_INTERPOLATED_PART_A_12345;
+			const originalPartB = process.env.TEST_INTERPOLATED_PART_B_12345;
+			process.env.TEST_INTERPOLATED_PART_A_12345 = "left";
+			process.env.TEST_INTERPOLATED_PART_B_12345 = "right";
+			const interpolatedKey = ["$", "{TEST_INTERPOLATED_PART_A_12345}_$", "{TEST_INTERPOLATED_PART_B_12345}"].join(
+				"",
+			);
+
+			try {
+				writeRawModelsJson({
+					"custom-provider": providerWithApiKey(interpolatedKey),
+				});
+
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+				const apiKey = await registry.getApiKeyForProvider("custom-provider");
+
+				expect(apiKey).toBe("left_right");
+			} finally {
+				if (originalPartA === undefined) {
+					delete process.env.TEST_INTERPOLATED_PART_A_12345;
+				} else {
+					process.env.TEST_INTERPOLATED_PART_A_12345 = originalPartA;
+				}
+				if (originalPartB === undefined) {
+					delete process.env.TEST_INTERPOLATED_PART_B_12345;
+				} else {
+					process.env.TEST_INTERPOLATED_PART_B_12345 = originalPartB;
+				}
+			}
+		});
+
+		test("apiKey with $$ prefix escapes a leading dollar", async () => {
+			writeRawModelsJson({
+				"custom-provider": providerWithApiKey("$$TEST_API_KEY_12345"),
+			});
+
+			const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+			const apiKey = await registry.getApiKeyForProvider("custom-provider");
+
+			expect(apiKey).toBe("$TEST_API_KEY_12345");
+		});
+
+		test("apiKey with $! escapes a literal bang and still interpolates later env refs", async () => {
+			const originalEnv = process.env.TEST_API_KEY_12345;
+			process.env.TEST_API_KEY_12345 = "env-api-key-value";
+
+			try {
+				writeRawModelsJson({
+					"custom-provider": providerWithApiKey("$!literal-$TEST_API_KEY_12345"),
+				});
+
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+				const apiKey = await registry.getApiKeyForProvider("custom-provider");
+
+				expect(apiKey).toBe("!literal-env-api-key-value");
+			} finally {
+				if (originalEnv === undefined) {
+					delete process.env.TEST_API_KEY_12345;
+				} else {
+					process.env.TEST_API_KEY_12345 = originalEnv;
+				}
+			}
+		});
+
+		test("plain apiKey is used directly even when it matches an env var", async () => {
 			const originalEnv = process.env.TEST_API_KEY_12345;
 			process.env.TEST_API_KEY_12345 = "env-api-key-value";
 
@@ -1199,7 +1432,7 @@ describe("ModelRegistry", () => {
 				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
 				const apiKey = await registry.getApiKeyForProvider("custom-provider");
 
-				expect(apiKey).toBe("env-api-key-value");
+				expect(apiKey).toBe("TEST_API_KEY_12345");
 			} finally {
 				if (originalEnv === undefined) {
 					delete process.env.TEST_API_KEY_12345;
@@ -1318,7 +1551,7 @@ describe("ModelRegistry", () => {
 					process.env[envVarName] = "status-test-key";
 
 					writeRawModelsJson({
-						"custom-provider": providerWithApiKey(envVarName),
+						"custom-provider": providerWithApiKey(`$${envVarName}`),
 					});
 
 					const registry = ModelRegistry.create(authStorage, modelsJsonPath);
@@ -1337,6 +1570,41 @@ describe("ModelRegistry", () => {
 				}
 			});
 
+			test("provider auth status reports interpolated apiKey environment variables", () => {
+				const envVarNameA = "TEST_API_KEY_STATUS_PART_A_98765";
+				const envVarNameB = "TEST_API_KEY_STATUS_PART_B_98765";
+				const originalEnvA = process.env[envVarNameA];
+				const originalEnvB = process.env[envVarNameB];
+				process.env[envVarNameA] = "left";
+				process.env[envVarNameB] = "right";
+				const interpolatedKey = ["$", "{", envVarNameA, "}_$", "{", envVarNameB, "}"].join("");
+
+				try {
+					writeRawModelsJson({
+						"custom-provider": providerWithApiKey(interpolatedKey),
+					});
+
+					const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+
+					expect(registry.getProviderAuthStatus("custom-provider")).toEqual({
+						configured: true,
+						source: "environment",
+						label: `${envVarNameA}, ${envVarNameB}`,
+					});
+				} finally {
+					if (originalEnvA === undefined) {
+						delete process.env[envVarNameA];
+					} else {
+						process.env[envVarNameA] = originalEnvA;
+					}
+					if (originalEnvB === undefined) {
+						delete process.env[envVarNameB];
+					} else {
+						process.env[envVarNameB] = originalEnvB;
+					}
+				}
+			});
+
 			test("provider auth status reports non-env apiKey values from models.json as a config key", () => {
 				writeRawModelsJson({
 					"custom-provider": providerWithApiKey("literal_api_key_value"),
@@ -1350,6 +1618,29 @@ describe("ModelRegistry", () => {
 				});
 			});
 
+			test("missing explicit env apiKey keeps provider unavailable", () => {
+				const envVarName = "TEST_API_KEY_MISSING_TEST_98765";
+				const originalEnv = process.env[envVarName];
+				delete process.env[envVarName];
+
+				try {
+					writeRawModelsJson({
+						"custom-provider": providerWithApiKey(`$${envVarName}`),
+					});
+
+					const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+
+					expect(registry.getProviderAuthStatus("custom-provider")).toEqual({ configured: false });
+					expect(registry.getAvailable().some((model) => model.provider === "custom-provider")).toBe(false);
+				} finally {
+					if (originalEnv === undefined) {
+						delete process.env[envVarName];
+					} else {
+						process.env[envVarName] = originalEnv;
+					}
+				}
+			});
+
 			test("provider auth status reports command apiKey values from models.json without executing them", () => {
 				const counterFile = join(tempDir, "status-counter");
 				writeFileSync(counterFile, "0");
@@ -1376,7 +1667,7 @@ describe("ModelRegistry", () => {
 					process.env[envVarName] = "first-value";
 
 					writeRawModelsJson({
-						"custom-provider": providerWithApiKey(envVarName),
+						"custom-provider": providerWithApiKey(`$${envVarName}`),
 					});
 
 					const registry = ModelRegistry.create(authStorage, modelsJsonPath);
@@ -1415,6 +1706,25 @@ describe("ModelRegistry", () => {
 				expect(count).toBe(0);
 			});
 
+			test("getAvailable filters GitHub Copilot OAuth models to account picker availability", () => {
+				authStorage.set("github-copilot", {
+					type: "oauth",
+					refresh: "github-access-token",
+					access: "tid=test;exp=9999999999;proxy-ep=proxy.individual.githubcopilot.com;",
+					expires: Date.now() + 60_000,
+					availableModelIds: ["gpt-4.1"],
+				});
+
+				const registry = ModelRegistry.create(authStorage, modelsJsonPath);
+
+				expect(
+					registry
+						.getAvailable()
+						.filter((m) => m.provider === "github-copilot")
+						.map((m) => m.id),
+				).toEqual(["gpt-4.1"]);
+			});
+
 			test("getApiKeyAndHeaders resolves authHeader on every request", async () => {
 				const tokenFile = join(tempDir, "token");
 				writeFileSync(tokenFile, "token-1");

packages/coding-agent/test/model-resolver.test.ts

@@ -344,6 +344,7 @@ describe("resolveCliModel", () => {
 		};
 		const registry = {
 			getAll: () => [...allModels, zaiModel, gatewayModel],
+			hasConfiguredAuth: () => true,
 		} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
 
 		const result = resolveCliModel({
@@ -356,6 +357,46 @@ describe("resolveCliModel", () => {
 		expect(result.model?.id).toBe("glm-5");
 	});
 
+	test("prefers an authenticated exact raw model id over an unauthenticated inferred provider", () => {
+		const commandcodeModel: Model<"anthropic-messages"> = {
+			id: "xiaomi/mimo-v2.5-pro",
+			name: "Xiaomi MiMo via Commandcode",
+			api: "anthropic-messages",
+			provider: "commandcode",
+			baseUrl: "https://example.invalid",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 1, output: 2, cacheRead: 0.1, cacheWrite: 1 },
+			contextWindow: 128000,
+			maxTokens: 8192,
+		};
+		const xiaomiModel: Model<"anthropic-messages"> = {
+			id: "mimo-v2.5-pro",
+			name: "Xiaomi MiMo",
+			api: "anthropic-messages",
+			provider: "xiaomi",
+			baseUrl: "https://api.xiaomimimo.com",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 1, output: 2, cacheRead: 0.1, cacheWrite: 1 },
+			contextWindow: 128000,
+			maxTokens: 8192,
+		};
+		const registry = {
+			getAll: () => [...allModels, commandcodeModel, xiaomiModel],
+			hasConfiguredAuth: (model: Model<"anthropic-messages">) => model.provider === "commandcode",
+		} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+		const result = resolveCliModel({
+			cliModel: "xiaomi/mimo-v2.5-pro",
+			modelRegistry: registry,
+		});
+
+		expect(result.error).toBeUndefined();
+		expect(result.model?.provider).toBe("commandcode");
+		expect(result.model?.id).toBe("xiaomi/mimo-v2.5-pro");
+	});
+
 	test("resolves provider-prefixed fuzzy patterns (openrouter/qwen -> openrouter model)", () => {
 		const registry = {
 			getAll: () => allModels,
@@ -370,6 +411,128 @@ describe("resolveCliModel", () => {
 		expect(result.model?.provider).toBe("openrouter");
 		expect(result.model?.id).toBe("qwen/qwen3-coder:exacto");
 	});
+
+	describe("custom model fallback with :thinking suffix (#5552)", () => {
+		// Models for a provider that has registered models but the specific model ID
+		// is not in the registry (triggers buildFallbackModel path).
+		const neuralwattModel: Model<"anthropic-messages"> = {
+			id: "some-base-model",
+			name: "Some Base Model",
+			api: "anthropic-messages",
+			provider: "neuralwatt",
+			baseUrl: "https://api.neuralwatt.com",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 1, output: 2, cacheRead: 0.1, cacheWrite: 1 },
+			contextWindow: 128000,
+			maxTokens: 8192,
+		};
+
+		const modelsWithNeuralwatt = [...allModels, neuralwattModel];
+
+		test("strips :thinking suffix from custom model id in fallback path", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			const result = resolveCliModel({
+				cliModel: "neuralwatt/zai-org/GLM-5.1-FP8:high",
+				modelRegistry: registry,
+			});
+
+			expect(result.error).toBeUndefined();
+			expect(result.model?.provider).toBe("neuralwatt");
+			// The :high suffix must NOT leak into the model id sent to the API
+			expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8");
+			expect(result.model?.reasoning).toBe(true);
+			expect(result.thinkingLevel).toBe("high");
+		});
+
+		test("custom model without thinking suffix works normally in fallback path", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			const result = resolveCliModel({
+				cliModel: "neuralwatt/zai-org/GLM-5.1-FP8",
+				modelRegistry: registry,
+			});
+
+			expect(result.error).toBeUndefined();
+			expect(result.model?.provider).toBe("neuralwatt");
+			expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8");
+			expect(result.thinkingLevel).toBeUndefined();
+		});
+
+		test("all valid thinking levels work in fallback path", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			for (const level of ["off", "minimal", "low", "medium", "high", "xhigh"]) {
+				const result = resolveCliModel({
+					cliModel: `neuralwatt/zai-org/GLM-5.1-FP8:${level}`,
+					modelRegistry: registry,
+				});
+
+				expect(result.error).toBeUndefined();
+				expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8");
+				expect(result.thinkingLevel).toBe(level);
+			}
+		});
+
+		test("invalid thinking suffix on custom model is treated as part of model id", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			const result = resolveCliModel({
+				cliModel: "neuralwatt/zai-org/GLM-5.1-FP8:banana",
+				modelRegistry: registry,
+			});
+
+			expect(result.error).toBeUndefined();
+			expect(result.model?.provider).toBe("neuralwatt");
+			// Invalid suffix stays in the id (it's not a thinking level)
+			expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8:banana");
+			expect(result.thinkingLevel).toBeUndefined();
+		});
+
+		test("explicit --provider with custom model:thinking strips suffix correctly", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			const result = resolveCliModel({
+				cliProvider: "neuralwatt",
+				cliModel: "zai-org/GLM-5.1-FP8:high",
+				modelRegistry: registry,
+			});
+
+			expect(result.error).toBeUndefined();
+			expect(result.model?.provider).toBe("neuralwatt");
+			expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8");
+			expect(result.thinkingLevel).toBe("high");
+		});
+
+		test("with explicit --thinking, :suffix is kept as part of model id", () => {
+			const registry = {
+				getAll: () => modelsWithNeuralwatt,
+			} as unknown as Parameters<typeof resolveCliModel>[0]["modelRegistry"];
+
+			const result = resolveCliModel({
+				cliModel: "neuralwatt/zai-org/GLM-5.1-FP8:high",
+				cliThinking: "medium",
+				modelRegistry: registry,
+			});
+
+			expect(result.error).toBeUndefined();
+			expect(result.model?.provider).toBe("neuralwatt");
+			// :high is kept as part of the model id since --thinking was explicit
+			expect(result.model?.id).toBe("zai-org/GLM-5.1-FP8:high");
+			expect(result.thinkingLevel).toBeUndefined();
+		});
+	});
 });
 
 describe("default model selection", () => {
@@ -378,11 +541,12 @@ describe("default model selection", () => {
 		expect(defaultModelPerProvider["openai-codex"]).toBe("gpt-5.5");
 	});
 
-	test("zai, minimax, and cerebras defaults track current models", () => {
+	test("zai, minimax, cerebras, and ant-ling defaults track current models", () => {
 		expect(defaultModelPerProvider.zai).toBe("glm-5.1");
 		expect(defaultModelPerProvider.minimax).toBe("MiniMax-M2.7");
 		expect(defaultModelPerProvider["minimax-cn"]).toBe("MiniMax-M2.7");
 		expect(defaultModelPerProvider.cerebras).toBe("zai-glm-4.7");
+		expect(defaultModelPerProvider["ant-ling"]).toBe("Ring-2.6-1T");
 	});
 
 	test("ai-gateway default tracks current model", () => {

packages/coding-agent/test/package-command-paths.test.ts

@@ -1,9 +1,11 @@
-import { mkdirSync, readFileSync, realpathSync, rmSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, realpathSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { ENV_AGENT_DIR, PACKAGE_NAME, VERSION } from "../src/config.ts";
+import { ProjectTrustStore } from "../src/core/trust-manager.ts";
 import { main } from "../src/main.ts";
+import { handlePackageCommand } from "../src/package-manager-cli.ts";
 
 describe("package commands", () => {
 	let tempDir: string;
@@ -21,6 +23,10 @@ describe("package commands", () => {
 		return `${major}.${minor}.${Number.parseInt(patch, 10) + 1}`;
 	}
 
+	async function runPackageCommandDirectly(args: string[]): Promise<void> {
+		expect(await handlePackageCommand(args)).toBe(true);
+	}
+
 	beforeEach(() => {
 		tempDir = join(tmpdir(), `pi-package-commands-${Date.now()}-${Math.random().toString(36).slice(2)}`);
 		agentDir = join(tempDir, "agent");
@@ -36,12 +42,21 @@ describe("package commands", () => {
 		originalExitCode = process.exitCode;
 		originalExecPath = process.execPath;
 		process.exitCode = undefined;
+		vi.spyOn(process, "exit").mockImplementation(((code?: string | number | null) => {
+			if (code === undefined || code === null || Number(code) === 0) {
+				process.exitCode = undefined;
+			} else {
+				process.exitCode = code;
+			}
+			return undefined as never;
+		}) as typeof process.exit);
 		process.env[ENV_AGENT_DIR] = agentDir;
 		process.chdir(projectDir);
 	});
 
 	afterEach(() => {
 		vi.unstubAllGlobals();
+		vi.restoreAllMocks();
 		process.chdir(originalCwd);
 		process.exitCode = originalExitCode;
 		if (originalAgentDir === undefined) {
@@ -85,6 +100,231 @@ describe("package commands", () => {
 		expect(removedSettings.packages ?? []).toHaveLength(0);
 	});
 
+	it("skips untrusted project package settings", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("No packages installed.");
+			expect(stdout).not.toContain("Project packages:");
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("uses remembered project trust for list", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		new ProjectTrustStore(agentDir).set(projectDir, true);
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("Project packages:");
+			expect(stdout).toContain("npm:@project/pkg");
+			expect(stdout).not.toContain("No packages installed.");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("overrides remembered trust for list with --no-approve", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		new ProjectTrustStore(agentDir).set(projectDir, true);
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list", "--no-approve"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("No packages installed.");
+			expect(stdout).not.toContain("Project packages:");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("approves project trust for list with --approve", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list", "--approve"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("Project packages:");
+			expect(stdout).toContain("npm:@project/pkg");
+			expect(stdout).not.toContain("No packages installed.");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("uses default project trust for list", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ defaultProjectTrust: "always" }));
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("Project packages:");
+			expect(stdout).toContain("npm:@project/pkg");
+			expect(stdout).not.toContain("No packages installed.");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("uses project_trust extensions for package commands", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(
+				main(["list"], {
+					extensionFactories: [
+						(pi) => {
+							pi.on("project_trust", () => ({ trusted: "yes" }));
+						},
+					],
+				}),
+			).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("Project packages:");
+			expect(stdout).toContain("npm:@project/pkg");
+			expect(stdout).not.toContain("No packages installed.");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("does not prompt or ask extensions for project trust during update", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ defaultProjectTrust: "always" }));
+		const fakeNpmPath = join(tempDir, "fake-project-npm.cjs");
+		const recordPath = join(tempDir, "project-update.json");
+		writeFileSync(
+			fakeNpmPath,
+			`const fs=require("node:fs");fs.writeFileSync(${JSON.stringify(recordPath)},JSON.stringify(process.argv.slice(2)));`,
+		);
+		writeFileSync(
+			join(projectDir, ".pi", "settings.json"),
+			JSON.stringify({ packages: ["npm:fake-package"], npmCommand: [originalExecPath, fakeNpmPath] }),
+		);
+		let projectTrustCalled = false;
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(
+				main(["update", "--extensions"], {
+					extensionFactories: [
+						(pi) => {
+							pi.on("project_trust", () => {
+								projectTrustCalled = true;
+								return { trusted: "yes" };
+							});
+						},
+					],
+				}),
+			).resolves.toBeUndefined();
+
+			expect(projectTrustCalled).toBe(false);
+			expect(existsSync(recordPath)).toBe(false);
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("uses saved project trust during update", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		const fakeNpmPath = join(tempDir, "fake-trusted-project-npm.cjs");
+		const recordPath = join(tempDir, "trusted-project-update.json");
+		writeFileSync(
+			fakeNpmPath,
+			`const fs=require("node:fs");fs.writeFileSync(${JSON.stringify(recordPath)},JSON.stringify(process.argv.slice(2)));`,
+		);
+		writeFileSync(
+			join(projectDir, ".pi", "settings.json"),
+			JSON.stringify({ packages: ["npm:fake-package"], npmCommand: [originalExecPath, fakeNpmPath] }),
+		);
+		new ProjectTrustStore(agentDir).set(projectDir, true);
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["update", "--extensions"])).resolves.toBeUndefined();
+
+			expect(existsSync(recordPath)).toBe(true);
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("lets trust.json override default project trust", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ defaultProjectTrust: "always" }));
+		writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ packages: ["npm:@project/pkg"] }));
+		new ProjectTrustStore(agentDir).set(projectDir, false);
+		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+		try {
+			await expect(main(["list"])).resolves.toBeUndefined();
+
+			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stdout).toContain("No packages installed.");
+			expect(stdout).not.toContain("Project packages:");
+			expect(process.exitCode).toBeUndefined();
+		} finally {
+			logSpy.mockRestore();
+		}
+	});
+
+	it("blocks local package changes when project is untrusted", async () => {
+		mkdirSync(join(projectDir, ".pi"), { recursive: true });
+		writeFileSync(join(projectDir, ".pi", "settings.json"), "{}");
+		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
+
+		try {
+			await expect(main(["install", "-l", "./local-package"])).resolves.toBeUndefined();
+
+			const stderr = errorSpy.mock.calls.map(([message]) => String(message)).join("\n");
+			expect(stderr).toContain("Project is not trusted. Use --approve to modify local package config.");
+			expect(process.exitCode).toBe(1);
+		} finally {
+			errorSpy.mockRestore();
+		}
+	});
+
+	it("allows local package install to initialize fresh project settings", async () => {
+		await main(["install", "-l", packageDir]);
+
+		const settingsPath = join(projectDir, ".pi", "settings.json");
+		const settings = JSON.parse(readFileSync(settingsPath, "utf-8")) as { packages?: string[] };
+		expect(settings.packages?.length).toBe(1);
+		const stored = settings.packages?.[0] ?? "";
+		expect(realpathSync(join(projectDir, ".pi", stored))).toBe(realpathSync(packageDir));
+		expect(process.exitCode).toBeUndefined();
+	});
+
 	it("shows install subcommand help", async () => {
 		const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
 		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
@@ -111,7 +351,7 @@ describe("package commands", () => {
 
 			const stderr = errorSpy.mock.calls.map(([message]) => String(message)).join("\n");
 			expect(stderr).toContain('Unknown option --unknown for "install".');
-			expect(stderr).toContain('Use "pi --help" or "pi install <source> [-l]".');
+			expect(stderr).toContain('Use "pi --help" or "pi install <source> [-l] [--approve|--no-approve]".');
 			expect(process.exitCode).toBe(1);
 		} finally {
 			errorSpy.mockRestore();
@@ -169,7 +409,7 @@ else fs.writeFileSync(${JSON.stringify(recordPath)},JSON.stringify(args));
 		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
 
 		try {
-			await expect(main(["update", "--self", "--force"])).resolves.toBeUndefined();
+			await expect(runPackageCommandDirectly(["update", "--self", "--force"])).resolves.toBeUndefined();
 
 			expect(process.exitCode).toBeUndefined();
 			expect(errorSpy).not.toHaveBeenCalled();
@@ -213,7 +453,7 @@ else fs.writeFileSync(${JSON.stringify(recordPath)},JSON.stringify(args));
 		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
 
 		try {
-			await expect(main(["update", "--self"])).resolves.toBeUndefined();
+			await expect(runPackageCommandDirectly(["update", "--self"])).resolves.toBeUndefined();
 
 			expect(process.exitCode).toBeUndefined();
 			expect(errorSpy).not.toHaveBeenCalled();
@@ -262,7 +502,7 @@ else {
 		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
 
 		try {
-			await expect(main(["update", "--self"])).resolves.toBeUndefined();
+			await expect(runPackageCommandDirectly(["update", "--self"])).resolves.toBeUndefined();
 
 			expect(process.exitCode).toBeUndefined();
 			expect(errorSpy).not.toHaveBeenCalled();
@@ -315,7 +555,7 @@ if(args.includes("install")) process.exit(23);
 		const errorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
 
 		try {
-			await expect(main(["update", "--self"])).resolves.toBeUndefined();
+			await expect(runPackageCommandDirectly(["update", "--self"])).resolves.toBeUndefined();
 
 			expect(process.exitCode).toBe(1);
 			const stdout = logSpy.mock.calls.map(([message]) => String(message)).join("\n");

packages/coding-agent/test/package-manager.test.ts

@@ -1,5 +1,5 @@
 import { EventEmitter } from "node:events";
-import { mkdirSync, rmSync, symlinkSync, writeFileSync } from "node:fs";
+import { mkdirSync, rmSync, statSync, symlinkSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join, relative } from "node:path";
 import { PassThrough } from "node:stream";
@@ -33,6 +33,20 @@ interface PackageManagerInternals {
 		options?: { cwd?: string; timeoutMs?: number; env?: Record<string, string> },
 	): Promise<string>;
 	getLocalGitUpdateTarget(installedPath: string): Promise<{ ref: string; head: string; fetchArgs: string[] }>;
+	parseSource(
+		source: string,
+	):
+		| { type: "npm"; spec: string; name: string; pinned: boolean }
+		| { type: "git"; repo: string; host: string; path: string; pinned: boolean; ref?: string }
+		| { type: "local"; path: string };
+	getNpmInstallPath(
+		source: { type: "npm"; spec: string; name: string; pinned: boolean },
+		scope: "user" | "project" | "temporary",
+	): string;
+	getGitInstallPath(
+		source: { type: "git"; repo: string; host: string; path: string; pinned: boolean; ref?: string },
+		scope: "user" | "project" | "temporary",
+	): string;
 }
 
 // Helper to check if a resource is enabled
@@ -693,7 +707,17 @@ Content`,
 
 			expect(runCommandSpy).toHaveBeenCalledWith(
 				"mise",
-				["exec", "node@20", "--", "npm", "install", "@scope/pkg", "--prefix", join(agentDir, "npm")],
+				[
+					"exec",
+					"node@20",
+					"--",
+					"npm",
+					"install",
+					"@scope/pkg",
+					"--prefix",
+					join(agentDir, "npm"),
+					"--legacy-peer-deps",
+				],
 				undefined,
 			);
 		});
@@ -714,7 +738,7 @@ Content`,
 
 			expect(runCommandSpy).toHaveBeenCalledWith(
 				"mise",
-				["exec", "bun@1", "--", "bun", "install", "@scope/pkg", "--cwd", join(agentDir, "npm")],
+				["exec", "bun@1", "--", "bun", "install", "@scope/pkg", "--cwd", join(agentDir, "npm"), "--omit=peer"],
 				undefined,
 			);
 		});
@@ -748,7 +772,7 @@ Content`,
 				if (args[0] === "rev-parse" && args[1] === "HEAD") {
 					return "old-head";
 				}
-				if (args[0] === "rev-parse" && args[1] === "FETCH_HEAD") {
+				if (args[0] === "rev-parse" && args[1] === "FETCH_HEAD^{commit}") {
 					return "new-head";
 				}
 				throw new Error(`Unexpected runCommandCapture args: ${args.join(" ")}`);
@@ -758,7 +782,9 @@ Content`,
 			await packageManager.install(source);
 
 			expect(runCommandSpy).toHaveBeenCalledWith("git", ["fetch", "origin", "v2"], { cwd: targetDir });
-			expect(runCommandSpy).toHaveBeenCalledWith("git", ["reset", "--hard", "FETCH_HEAD"], { cwd: targetDir });
+			expect(runCommandSpy).toHaveBeenCalledWith("git", ["reset", "--hard", "FETCH_HEAD^{commit}"], {
+				cwd: targetDir,
+			});
 			expect(runCommandSpy).toHaveBeenCalledWith("git", ["clean", "-fdx"], { cwd: targetDir });
 			expect(runCommandSpy).toHaveBeenCalledWith("npm", ["install", "--omit=dev"], { cwd: targetDir });
 		});
@@ -779,7 +805,7 @@ Content`,
 				if (args[0] === "rev-parse" && args[1] === "HEAD") {
 					return "old-head";
 				}
-				if (args[0] === "rev-parse" && args[1] === "origin/HEAD") {
+				if (args[0] === "rev-parse" && args[1] === "origin/HEAD^{commit}") {
 					return "new-head";
 				}
 				throw new Error(`Unexpected runCommandCapture args: ${args.join(" ")}`);
@@ -789,7 +815,9 @@ Content`,
 			await packageManager.install(source);
 
 			expect(runCommandSpy).toHaveBeenCalledWith("git", fetchArgs, { cwd: targetDir });
-			expect(runCommandSpy).toHaveBeenCalledWith("git", ["reset", "--hard", "origin/HEAD"], { cwd: targetDir });
+			expect(runCommandSpy).toHaveBeenCalledWith("git", ["reset", "--hard", "origin/HEAD^{commit}"], {
+				cwd: targetDir,
+			});
 			expect(runCommandSpy).toHaveBeenCalledWith("git", ["clean", "-fdx"], { cwd: targetDir });
 		});
 
@@ -832,7 +860,7 @@ Content`,
 				if (args[0] === "rev-parse" && args[1] === "--abbrev-ref" && args[2] === "@{upstream}") {
 					return "origin/main";
 				}
-				if (args[0] === "rev-parse" && args[1] === "@{upstream}") {
+				if (args[0] === "rev-parse" && (args[1] === "@{upstream}" || args[1] === "@{upstream}^{commit}")) {
 					return "remote-head";
 				}
 				if (args[0] === "rev-parse" && args[1] === "HEAD") {
@@ -868,7 +896,7 @@ Content`,
 				if (args[0] === "rev-parse" && args[1] === "--abbrev-ref" && args[2] === "@{upstream}") {
 					return "origin/main";
 				}
-				if (args[0] === "rev-parse" && args[1] === "@{upstream}") {
+				if (args[0] === "rev-parse" && (args[1] === "@{upstream}" || args[1] === "@{upstream}^{commit}")) {
 					return "remote-head";
 				}
 				if (args[0] === "rev-parse" && args[1] === "HEAD") {
@@ -949,6 +977,8 @@ Content`,
 						"pnpm-pkg",
 						"--prefix",
 						join(agentDir, "npm"),
+						"--config.auto-install-peers=false",
+						"--config.strict-peer-dependencies=false",
 						"--config.strict-dep-builds=false",
 					]);
 					mkdirSync(join(packagePath, "extensions"), { recursive: true });
@@ -1098,8 +1128,17 @@ Content`,
 		});
 
 		it("should parse package source types from docs examples", () => {
-			expect((packageManager as any).parseSource("npm:@scope/pkg@1.2.3").type).toBe("npm");
-			expect((packageManager as any).parseSource("npm:pkg").type).toBe("npm");
+			const parseNpm = (source: string) => {
+				const parsed = (packageManager as any).parseSource(source);
+				if (parsed.type !== "npm") {
+					throw new Error(`Expected npm source: ${source}`);
+				}
+				return parsed;
+			};
+
+			expect(parseNpm("npm:@scope/pkg@1.2.3").pinned).toBe(true);
+			expect(parseNpm("npm:@scope/pkg@^1.2.3").pinned).toBe(false);
+			expect(parseNpm("npm:pkg").pinned).toBe(false);
 
 			expect((packageManager as any).parseSource("git:github.com/user/repo@v1").type).toBe("git");
 			expect((packageManager as any).parseSource("https://github.com/user/repo@v1").type).toBe("git");
@@ -1122,6 +1161,45 @@ Content`,
 		});
 	});
 
+	describe("git install paths", () => {
+		it("should reject paths outside git install roots", () => {
+			const managerWithInternals = packageManager as unknown as PackageManagerInternals;
+			const traversalSource = {
+				type: "git" as const,
+				repo: "git@evil.example:../../victim/repo",
+				host: "evil.example",
+				path: "../../victim/repo",
+				pinned: false,
+			};
+
+			for (const scope of ["user", "project", "temporary"] as const) {
+				expect(() => managerWithInternals.getGitInstallPath(traversalSource, scope)).toThrow(
+					"outside package install root",
+				);
+			}
+		});
+	});
+
+	describe("temporary install paths", () => {
+		it("should place temporary npm packages under the agent temp extension folder", () => {
+			const managerWithInternals = packageManager as unknown as PackageManagerInternals;
+			const source = managerWithInternals.parseSource("npm:left-pad");
+			if (source.type !== "npm") {
+				throw new Error("Expected npm source");
+			}
+
+			const installPath = managerWithInternals.getNpmInstallPath(source, "temporary");
+			const tempRoot = join(agentDir, "tmp", "extensions");
+
+			expect(pathEndsWith(installPath, "node_modules/left-pad")).toBe(true);
+			expect(relative(tempRoot, installPath).startsWith("..")).toBe(false);
+			expect(installPath.startsWith(join(tmpdir(), "pi-extensions"))).toBe(false);
+			if (process.platform !== "win32") {
+				expect(statSync(tempRoot).mode & 0o777).toBe(0o700);
+			}
+		});
+	});
+
 	describe("settings source normalization", () => {
 		it("should store global local packages relative to agent settings base", () => {
 			const pkgDir = join(tempDir, "packages", "local-global-pkg");
@@ -1983,25 +2061,27 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 	});
 
 	describe("offline mode and network timeouts", () => {
-		it("should update project npm packages using @latest when newer version is available", async () => {
+		it("should update npm range packages using the configured spec", async () => {
 			const installedPath = join(tempDir, ".pi", "npm", "node_modules", "example");
 			mkdirSync(installedPath, { recursive: true });
 			writeFileSync(join(installedPath, "package.json"), JSON.stringify({ name: "example", version: "1.0.0" }));
-			settingsManager.setProjectPackages(["npm:example"]);
+			settingsManager.setProjectPackages(["npm:example@^1.0.0"]);
 
-			const runCommandCaptureSpy = vi.spyOn(packageManager as any, "runCommandCapture").mockResolvedValue('"1.2.3"');
+			const runCommandCaptureSpy = vi
+				.spyOn(packageManager as any, "runCommandCapture")
+				.mockResolvedValue('["1.0.0","1.2.0"]');
 			const runCommandSpy = vi.spyOn(packageManager as any, "runCommand").mockResolvedValue(undefined);
 
 			await packageManager.update("npm:example");
 
 			expect(runCommandCaptureSpy).toHaveBeenCalledWith(
 				"npm",
-				["view", "example", "version", "--json"],
+				["view", "example@^1.0.0", "version", "--json"],
 				expect.objectContaining({ cwd: tempDir, timeoutMs: expect.any(Number) }),
 			);
 			expect(runCommandSpy).toHaveBeenCalledWith(
 				"npm",
-				["install", "example@latest", "--prefix", join(tempDir, ".pi", "npm")],
+				["install", "example@^1.0.0", "--prefix", join(tempDir, ".pi", "npm"), "--legacy-peer-deps"],
 				undefined,
 			);
 		});
@@ -2009,17 +2089,19 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 		it("should skip project npm update when installed version matches latest", async () => {
 			const installedPath = join(tempDir, ".pi", "npm", "node_modules", "example");
 			mkdirSync(installedPath, { recursive: true });
-			writeFileSync(join(installedPath, "package.json"), JSON.stringify({ name: "example", version: "1.2.3" }));
-			settingsManager.setProjectPackages(["npm:example"]);
+			writeFileSync(join(installedPath, "package.json"), JSON.stringify({ name: "example", version: "1.3.1" }));
+			settingsManager.setProjectPackages(["npm:example@^1.0.0"]);
 
-			const runCommandCaptureSpy = vi.spyOn(packageManager as any, "runCommandCapture").mockResolvedValue('"1.2.3"');
+			const runCommandCaptureSpy = vi
+				.spyOn(packageManager as any, "runCommandCapture")
+				.mockResolvedValue('["1.0.0","1.3.1","1.0.2"]');
 			const runCommandSpy = vi.spyOn(packageManager as any, "runCommand").mockResolvedValue(undefined);
 
 			await packageManager.update("npm:example");
 
 			expect(runCommandCaptureSpy).toHaveBeenCalledWith(
 				"npm",
-				["view", "example", "version", "--json"],
+				["view", "example@^1.0.0", "version", "--json"],
 				expect.objectContaining({ cwd: tempDir, timeoutMs: expect.any(Number) }),
 			);
 			expect(runCommandSpy).not.toHaveBeenCalled();
@@ -2040,7 +2122,13 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 				.mockImplementation(async (...callArgs: unknown[]) => {
 					const [command, args] = callArgs as [string, string[]];
 					expect(command).toBe("npm");
-					expect(args).toEqual(["install", "legacy-pkg@latest", "--prefix", join(agentDir, "npm")]);
+					expect(args).toEqual([
+						"install",
+						"legacy-pkg@latest",
+						"--prefix",
+						join(agentDir, "npm"),
+						"--legacy-peer-deps",
+					]);
 					mkdirSync(managedPath, { recursive: true });
 					writeFileSync(
 						join(managedPath, "package.json"),
@@ -2057,7 +2145,7 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 			expect(packageManager.getInstalledPath("npm:legacy-pkg", "user")).toBe(managedPath);
 		});
 
-		it("should batch npm updates per scope and run git updates in parallel while skipping pinned and current packages", async () => {
+		it("should batch npm updates per scope and run git updates in parallel while skipping pinned npm and current packages", async () => {
 			const userOldPath = join(agentDir, "npm", "node_modules", "user-old");
 			const userCurrentPath = join(agentDir, "npm", "node_modules", "user-current");
 			const userUnknownPath = join(agentDir, "npm", "node_modules", "user-unknown");
@@ -2150,16 +2238,30 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 			expect(runCommandSpy).toHaveBeenNthCalledWith(
 				1,
 				"npm",
-				["install", "user-old@latest", "user-unknown@latest", "--prefix", join(agentDir, "npm")],
+				[
+					"install",
+					"user-old@latest",
+					"user-unknown@latest",
+					"--prefix",
+					join(agentDir, "npm"),
+					"--legacy-peer-deps",
+				],
 				undefined,
 			);
 			expect(runCommandSpy).toHaveBeenNthCalledWith(
 				2,
 				"npm",
-				["install", "project-old@latest", "project-missing@latest", "--prefix", join(tempDir, ".pi", "npm")],
+				[
+					"install",
+					"project-old@latest",
+					"project-missing@latest",
+					"--prefix",
+					join(tempDir, ".pi", "npm"),
+					"--legacy-peer-deps",
+				],
 				undefined,
 			);
-			expect(updateGitSpy).toHaveBeenCalledTimes(3);
+			expect(updateGitSpy).toHaveBeenCalledTimes(4);
 			expect(maxConcurrentNpmUpdates).toBeGreaterThan(1);
 			expect(maxConcurrentGitUpdates).toBeGreaterThan(1);
 		});
@@ -2209,11 +2311,12 @@ export default function(api) { api.registerTool({ name: "test", description: "te
 		});
 
 		it("should not run npm view during resolve for installed unpinned packages", async () => {
+			process.env.PI_OFFLINE = "1";
 			const installedPath = join(tempDir, ".pi", "npm", "node_modules", "example");
 			mkdirSync(join(installedPath, "extensions"), { recursive: true });
 			writeFileSync(join(installedPath, "package.json"), JSON.stringify({ name: "example", version: "1.0.0" }));
 			writeFileSync(join(installedPath, "extensions", "index.ts"), "export default function() {};");
-			settingsManager.setProjectPackages(["npm:example"]);
+			settingsManager.setProjectPackages(["npm:example@^1.0.0"]);
 
 			const runCommandCaptureSpy = vi.spyOn(packageManager as any, "runCommandCapture");
 

packages/coding-agent/test/prompt-templates.test.ts

@@ -190,6 +190,52 @@ describe("substituteArgs", () => {
 	});
 });
 
+// ============================================================================
+// substituteArgs - Positional Defaults
+// ============================================================================
+
+describe("substituteArgs - positional defaults", () => {
+	test("should use default when positional arg is missing", () => {
+		expect(substituteArgs(`List exactly \${1:-7} next steps`, [])).toBe("List exactly 7 next steps");
+	});
+
+	test("should use positional arg when present", () => {
+		expect(substituteArgs(`List exactly \${1:-7} next steps`, ["3"])).toBe("List exactly 3 next steps");
+	});
+
+	test("should use default when positional arg is empty", () => {
+		expect(substituteArgs(`Mode: \${1:-brief}`, [""])).toBe("Mode: brief");
+	});
+
+	test("should support multiple positional defaults", () => {
+		expect(substituteArgs(`\${1:-7} \${2:-brief}`, [])).toBe("7 brief");
+		expect(substituteArgs(`\${1:-7} \${2:-brief}`, ["3"])).toBe("3 brief");
+		expect(substituteArgs(`\${1:-7} \${2:-brief}`, ["3", "verbose"])).toBe("3 verbose");
+	});
+
+	test("should not recursively substitute patterns in arg values", () => {
+		expect(substituteArgs(`\${1:-7}`, ["$ARGUMENTS"])).toBe("$ARGUMENTS");
+		expect(substituteArgs(`\${1:-7}`, ["$1"])).toBe("$1");
+	});
+
+	test("should not recursively substitute patterns in default values", () => {
+		expect(substituteArgs(`\${1:-$ARGUMENTS}`, ["a", "b"])).toBe("a");
+		expect(substituteArgs(`\${3:-$ARGUMENTS}`, ["a", "b"])).toBe("$ARGUMENTS");
+	});
+
+	test("should support defaults with spaces", () => {
+		expect(substituteArgs(`\${1:-seven steps}`, [])).toBe("seven steps");
+	});
+
+	test("should support out-of-range positional defaults", () => {
+		expect(substituteArgs(`\${3:-fallback}`, ["a", "b"])).toBe("fallback");
+	});
+
+	test("should mix positional defaults with existing placeholders", () => {
+		expect(substituteArgs(`$1 \${2:-x} $ARGUMENTS`, ["a"])).toBe("a x a");
+	});
+});
+
 // ============================================================================
 // substituteArgs - Array Slicing (Bash-Style)
 // ============================================================================

packages/coding-agent/test/resource-loader.test.ts

@@ -187,6 +187,53 @@ Project skill`,
 			expect(extensionsResult.extensions[0].path).toBe(join(cwd, ".pi", "extensions", "shared.ts"));
 		});
 
+		it("should load user extensions before trust and reuse them after trust resolves", async () => {
+			const userExtDir = join(agentDir, "extensions");
+			const projectExtDir = join(cwd, ".pi", "extensions");
+			mkdirSync(userExtDir, { recursive: true });
+			mkdirSync(projectExtDir, { recursive: true });
+			const loadCountKey = `__piTrustPreloadCount_${Date.now()}_${Math.random().toString(36).slice(2)}`;
+			const globalState = globalThis as typeof globalThis & Record<string, number | undefined>;
+
+			writeFileSync(
+				join(userExtDir, "user.ts"),
+				`globalThis[${JSON.stringify(loadCountKey)}] = (globalThis[${JSON.stringify(loadCountKey)}] ?? 0) + 1;
+export default function(pi) {
+	pi.on("project_trust", () => ({ trusted: "yes" }));
+	pi.registerCommand("user-trust", {
+		description: "user trust",
+		handler: async () => {},
+	});
+}`,
+			);
+			writeFileSync(
+				join(projectExtDir, "project.ts"),
+				`export default function(pi) {
+	pi.registerCommand("project-trusted", {
+		description: "project trusted",
+		handler: async () => {},
+	});
+}`,
+			);
+
+			const loader = new DefaultResourceLoader({ cwd, agentDir });
+			await loader.reload({
+				resolveProjectTrust: async ({ extensionsResult }) => {
+					expect(extensionsResult.extensions.map((extension) => extension.path)).toEqual([
+						join(userExtDir, "user.ts"),
+					]);
+					return true;
+				},
+			});
+
+			const extensionsResult = loader.getExtensions();
+			expect(extensionsResult.extensions.map((extension) => extension.path)).toEqual([
+				join(cwd, ".pi", "extensions", "project.ts"),
+				join(userExtDir, "user.ts"),
+			]);
+			expect(globalState[loadCountKey]).toBe(1);
+		});
+
 		it("should keep both extensions loaded when command names collide", async () => {
 			const userExtDir = join(agentDir, "extensions");
 			const projectExtDir = join(cwd, ".pi", "extensions");
@@ -329,6 +376,52 @@ Content`,
 			expect(loader.getSystemPrompt()).toBe("You are a helpful assistant.");
 		});
 
+		it("should skip project resources that require trust when project is not trusted", async () => {
+			const piDir = join(cwd, ".pi");
+			const extensionsDir = join(piDir, "extensions");
+			const skillDir = join(piDir, "skills", "project-skill");
+			const promptsDir = join(piDir, "prompts");
+			const themesDir = join(piDir, "themes");
+			mkdirSync(extensionsDir, { recursive: true });
+			mkdirSync(skillDir, { recursive: true });
+			mkdirSync(promptsDir, { recursive: true });
+			mkdirSync(themesDir, { recursive: true });
+			writeFileSync(join(piDir, "SYSTEM.md"), "Project system prompt.");
+			writeFileSync(join(agentDir, "SYSTEM.md"), "Global system prompt.");
+			writeFileSync(join(agentDir, "AGENTS.md"), "Global instructions");
+			writeFileSync(join(cwd, "AGENTS.md"), "Project instructions");
+			writeFileSync(join(extensionsDir, "project.ts"), `throw new Error("should not load");`);
+			writeFileSync(
+				join(skillDir, "SKILL.md"),
+				`---
+name: project-skill
+description: Project skill
+---
+Project skill content`,
+			);
+			writeFileSync(join(promptsDir, "project.md"), "Project prompt");
+			const themeData = JSON.parse(
+				readFileSync(join(process.cwd(), "src", "modes", "interactive", "theme", "dark.json"), "utf-8"),
+			) as { name: string };
+			themeData.name = "project-theme";
+			writeFileSync(join(themesDir, "project.json"), JSON.stringify(themeData, null, 2));
+			const settingsManager = SettingsManager.create(cwd, agentDir, { projectTrusted: false });
+
+			const loader = new DefaultResourceLoader({ cwd, agentDir, settingsManager });
+			await loader.reload();
+
+			expect(loader.getSystemPrompt()).toBe("Global system prompt.");
+			expect(loader.getAgentsFiles().agentsFiles.some((file) => file.path === join(agentDir, "AGENTS.md"))).toBe(
+				true,
+			);
+			expect(loader.getAgentsFiles().agentsFiles.some((file) => file.path === join(cwd, "AGENTS.md"))).toBe(true);
+			expect(loader.getExtensions().extensions).toHaveLength(0);
+			expect(loader.getExtensions().errors).toEqual([]);
+			expect(loader.getSkills().skills.some((skill) => skill.name === "project-skill")).toBe(false);
+			expect(loader.getPrompts().prompts.some((prompt) => prompt.name === "project")).toBe(false);
+			expect(loader.getThemes().themes.some((theme) => theme.name === "project-theme")).toBe(false);
+		});
+
 		it("should discover APPEND_SYSTEM.md", async () => {
 			const piDir = join(cwd, ".pi");
 			mkdirSync(piDir, { recursive: true });

packages/coding-agent/test/rpc-client-process-exit.test.ts

@@ -0,0 +1,38 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, test } from "vitest";
+import { RpcClient } from "../src/modes/rpc/rpc-client.ts";
+
+const tempDirs: string[] = [];
+
+function writeChildScript(contents: string): string {
+	const dir = mkdtempSync(join(tmpdir(), "pi-rpc-client-exit-"));
+	tempDirs.push(dir);
+	const path = join(dir, "child.mjs");
+	writeFileSync(path, contents);
+	return path;
+}
+
+afterEach(() => {
+	for (const dir of tempDirs.splice(0)) {
+		rmSync(dir, { recursive: true, force: true });
+	}
+});
+
+describe("RpcClient child process failures", () => {
+	test("rejects an in-flight request when the child process exits", async () => {
+		const client = new RpcClient({
+			cliPath: writeChildScript(`
+process.stdin.once("data", () => {
+	process.exit(43);
+});
+process.stdin.resume();
+`),
+		});
+
+		await client.start();
+
+		await expect(client.getCommands()).rejects.toThrow(/Agent process exited \(code=43 signal=null\)/);
+	});
+});

packages/coding-agent/test/rpc-prompt-response-semantics.test.ts

@@ -25,7 +25,9 @@ const rpcIo = vi.hoisted(() => ({
 }));
 
 vi.mock("../src/core/output-guard.js", () => ({
+	flushRawStdout: vi.fn(async () => {}),
 	takeOverStdout: vi.fn(),
+	waitForRawStdoutBackpressure: vi.fn(async () => {}),
 	writeRawStdout: (line: string) => {
 		rpcIo.outputLines.push(line);
 	},

packages/coding-agent/test/sdk-openrouter-attribution.test.ts

@@ -15,14 +15,14 @@ import { createAgentSession } from "../src/core/sdk.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 
-describe("createAgentSession OpenRouter attribution headers", () => {
+describe("createAgentSession provider attribution headers", () => {
 	let tempDir: string;
 	let cwd: string;
 	let agentDir: string;
 	let originalTelemetryEnv: string | undefined;
 
 	beforeEach(() => {
-		tempDir = join(tmpdir(), `pi-sdk-openrouter-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
+		tempDir = join(tmpdir(), `pi-sdk-attribution-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
 		cwd = join(tempDir, "project");
 		agentDir = join(tempDir, "agent");
 		mkdirSync(cwd, { recursive: true });
@@ -42,9 +42,9 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		}
 	});
 
-	function createModel(provider: string, baseUrl: string): Model<Api> {
+	function createModel(provider: string, baseUrl: string, id = `${provider}-test-model`): Model<Api> {
 		return {
-			id: `${provider}-test-model`,
+			id,
 			name: `${provider} Test Model`,
 			api: "openai-completions",
 			provider,
@@ -86,6 +86,7 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 			telemetryEnabled?: boolean;
 			providerHeaders?: Record<string, string>;
 			requestHeaders?: Record<string, string>;
+			sessionId?: string;
 		} = {},
 	): Promise<Record<string, string> | undefined> {
 		const settingsManager = SettingsManager.create(cwd, agentDir);
@@ -112,6 +113,11 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 			registeredProviders.push(model.provider);
 		}
 
+		const sessionManager = SessionManager.inMemory(cwd);
+		if (options.sessionId) {
+			sessionManager.newSession({ id: options.sessionId });
+		}
+
 		const { session } = await createAgentSession({
 			cwd,
 			agentDir,
@@ -119,14 +125,17 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 			authStorage,
 			modelRegistry,
 			settingsManager,
-			sessionManager: SessionManager.inMemory(cwd),
+			sessionManager,
 		});
 
 		try {
 			await session.agent.streamFn(
 				model,
 				{ messages: [] },
-				options.requestHeaders ? { headers: options.requestHeaders } : undefined,
+				{
+					sessionId: session.sessionId,
+					...(options.requestHeaders ? { headers: options.requestHeaders } : {}),
+				},
 			);
 			return capturedOptions?.headers;
 		} finally {
@@ -163,6 +172,14 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		expect(headers?.["X-OpenRouter-Categories"]).toBe("cli-agent");
 	});
 
+	it("preserves legacy OpenRouter base URL substring attribution matching", async () => {
+		const headers = await captureHeaders(createModel("custom-openrouter", "not-a-url-openrouter.ai"));
+
+		expect(headers?.["HTTP-Referer"]).toBe("https://pi.dev");
+		expect(headers?.["X-OpenRouter-Title"]).toBe("pi");
+		expect(headers?.["X-OpenRouter-Categories"]).toBe("cli-agent");
+	});
+
 	it("lets provider and request headers override the defaults", async () => {
 		const headers = await captureHeaders(createModel("openrouter", "https://openrouter.ai/api/v1"), {
 			providerHeaders: {
@@ -178,4 +195,83 @@ describe("createAgentSession OpenRouter attribution headers", () => {
 		expect(headers?.["X-OpenRouter-Title"]).toBe("request-title");
 		expect(headers?.["X-OpenRouter-Categories"]).toBe("provider-category");
 	});
+
+	it("adds default attribution headers for Vercel AI Gateway models", async () => {
+		const headers = await captureHeaders(createModel("vercel-ai-gateway", "https://ai-gateway.vercel.sh/v1"));
+
+		expect(headers?.["http-referer"]).toBe("https://pi.dev");
+		expect(headers?.["x-title"]).toBe("pi");
+	});
+
+	it("adds default attribution headers for direct NVIDIA NIM endpoints", async () => {
+		const headers = await captureHeaders(createModel("custom-nim", "https://integrate.api.nvidia.com/v1"));
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Pi");
+	});
+
+	it("adds default attribution headers for the NVIDIA provider", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://example.test/v1"));
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Pi");
+	});
+
+	it("does not add NVIDIA NIM attribution headers when telemetry is disabled", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://integrate.api.nvidia.com/v1"), {
+			telemetryEnabled: false,
+		});
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
+	it("lets provider and request headers override NVIDIA NIM defaults", async () => {
+		const headers = await captureHeaders(createModel("nvidia", "https://integrate.api.nvidia.com/v1"), {
+			providerHeaders: {
+				"X-BILLING-INVOKE-ORIGIN": "Provider",
+			},
+			requestHeaders: {
+				"X-BILLING-INVOKE-ORIGIN": "Request",
+			},
+		});
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBe("Request");
+	});
+
+	it("does not add NVIDIA NIM attribution headers for NVIDIA models routed through OpenRouter", async () => {
+		const headers = await captureHeaders(
+			createModel("openrouter", "https://openrouter.ai/api/v1", "nvidia/nemotron-3-super-120b-a12b"),
+		);
+
+		expect(headers?.["HTTP-Referer"]).toBe("https://pi.dev");
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
+	it("does not add NVIDIA NIM attribution headers for NVIDIA models routed through Vercel AI Gateway", async () => {
+		const headers = await captureHeaders(
+			createModel("vercel-ai-gateway", "https://ai-gateway.vercel.sh/v1", "nvidia/nemotron-3-super-120b-a12b"),
+		);
+
+		expect(headers?.["X-BILLING-INVOKE-ORIGIN"]).toBeUndefined();
+	});
+
+	it("adds OpenCode session headers", async () => {
+		const headers = await captureHeaders(createModel("opencode", "https://opencode.ai/zen/v1"), {
+			sessionId: "opencode-session",
+		});
+
+		expect(headers?.["x-opencode-session"]).toBe("opencode-session");
+		expect(headers?.["x-opencode-client"]).toBe("pi");
+	});
+
+	it("lets configured OpenCode headers override the defaults", async () => {
+		const headers = await captureHeaders(createModel("opencode", "https://opencode.ai/zen/v1"), {
+			sessionId: "opencode-session",
+			providerHeaders: {
+				"x-opencode-session": "configured-session",
+				"x-opencode-client": "configured-client",
+			},
+		});
+
+		expect(headers?.["x-opencode-session"]).toBe("configured-session");
+		expect(headers?.["x-opencode-client"]).toBe("configured-client");
+	});
 });

packages/coding-agent/test/sdk-stream-options.test.ts

@@ -0,0 +1,153 @@
+import { mkdirSync, mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import {
+	type Api,
+	type AssistantMessage,
+	createAssistantMessageEventStream,
+	type Model,
+	type SimpleStreamOptions,
+} from "@earendil-works/pi-ai";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { AuthStorage } from "../src/core/auth-storage.ts";
+import { ModelRegistry } from "../src/core/model-registry.ts";
+import { createAgentSession } from "../src/core/sdk.ts";
+import { SessionManager } from "../src/core/session-manager.ts";
+import { SettingsManager } from "../src/core/settings-manager.ts";
+
+describe("createAgentSession stream options", () => {
+	let tempDir: string;
+	let cwd: string;
+	let agentDir: string;
+
+	beforeEach(() => {
+		tempDir = mkdtempSync(join(tmpdir(), "pi-sdk-stream-options-"));
+		cwd = join(tempDir, "project");
+		agentDir = join(tempDir, "agent");
+		mkdirSync(cwd, { recursive: true });
+		mkdirSync(agentDir, { recursive: true });
+	});
+
+	afterEach(() => {
+		if (tempDir) {
+			rmSync(tempDir, { recursive: true, force: true });
+		}
+	});
+
+	function createModel(api: Api): Model<Api> {
+		return {
+			id: "capture-model",
+			name: "Capture Model",
+			api,
+			provider: "capture-provider",
+			baseUrl: "https://capture.invalid/v1",
+			reasoning: false,
+			input: ["text"],
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+			contextWindow: 128000,
+			maxTokens: 4096,
+		};
+	}
+
+	function createDoneStream(api: Api) {
+		const stream = createAssistantMessageEventStream();
+		const message: AssistantMessage = {
+			role: "assistant",
+			content: [{ type: "text", text: "ok" }],
+			api,
+			provider: "capture-provider",
+			model: "capture-model",
+			usage: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+				totalTokens: 0,
+				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+			},
+			stopReason: "stop",
+			timestamp: Date.now(),
+		};
+		stream.end(message);
+		return stream;
+	}
+
+	async function captureStreamOptions(
+		api: Api,
+		settings: { httpIdleTimeoutMs?: number; websocketConnectTimeoutMs?: number },
+		requestOptions: SimpleStreamOptions = {},
+	): Promise<SimpleStreamOptions | undefined> {
+		const model = createModel(api);
+		const settingsManager = SettingsManager.inMemory(settings);
+
+		const authStorage = AuthStorage.create(join(agentDir, "auth.json"));
+		authStorage.setRuntimeApiKey(model.provider, "test-api-key");
+		const modelRegistry = ModelRegistry.create(authStorage, join(agentDir, "models.json"));
+		let capturedOptions: SimpleStreamOptions | undefined;
+
+		modelRegistry.registerProvider(model.provider, {
+			api,
+			streamSimple: (_model, _context, providerOptions) => {
+				capturedOptions = providerOptions;
+				return createDoneStream(api);
+			},
+		});
+
+		const sessionManager = SessionManager.inMemory(cwd);
+		const { session } = await createAgentSession({
+			cwd,
+			agentDir,
+			model,
+			authStorage,
+			modelRegistry,
+			settingsManager,
+			sessionManager,
+		});
+
+		try {
+			await session.agent.streamFn(model, { messages: [] }, requestOptions);
+			return capturedOptions;
+		} finally {
+			session.dispose();
+			modelRegistry.unregisterProvider(model.provider);
+		}
+	}
+
+	it("forwards httpIdleTimeoutMs as timeoutMs for OpenAI Codex", async () => {
+		const options = await captureStreamOptions("openai-codex-responses", { httpIdleTimeoutMs: 1234 });
+
+		expect(options?.timeoutMs).toBe(1234);
+	});
+
+	it("defaults timeoutMs from httpIdleTimeoutMs for all providers", async () => {
+		const options = await captureStreamOptions("openai-completions", { httpIdleTimeoutMs: 1234 });
+
+		expect(options?.timeoutMs).toBe(1234);
+	});
+
+	it("lets request timeoutMs override httpIdleTimeoutMs for OpenAI Codex", async () => {
+		const options = await captureStreamOptions(
+			"openai-codex-responses",
+			{ httpIdleTimeoutMs: 1234 },
+			{ timeoutMs: 0 },
+		);
+
+		expect(options?.timeoutMs).toBe(0);
+	});
+
+	it("forwards websocketConnectTimeoutMs from settings", async () => {
+		const options = await captureStreamOptions("openai-codex-responses", { websocketConnectTimeoutMs: 1234 });
+
+		expect(options?.websocketConnectTimeoutMs).toBe(1234);
+	});
+
+	it("lets request websocketConnectTimeoutMs override settings", async () => {
+		const options = await captureStreamOptions(
+			"openai-codex-responses",
+			{ websocketConnectTimeoutMs: 1234 },
+			{ websocketConnectTimeoutMs: 0 },
+		);
+
+		expect(options?.websocketConnectTimeoutMs).toBe(0);
+	});
+});

packages/coding-agent/test/session-id-readonly.test.ts

@@ -0,0 +1,131 @@
+import { spawn } from "node:child_process";
+import { existsSync, mkdirSync, mkdtempSync, readdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join, resolve } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { ENV_AGENT_DIR } from "../src/config.ts";
+
+const cliPath = resolve(__dirname, "../src/cli.ts");
+const tempDirs: string[] = [];
+
+afterEach(() => {
+	for (const dir of tempDirs.splice(0)) {
+		rmSync(dir, { recursive: true, force: true });
+	}
+});
+
+function createTempDir(): string {
+	const dir = mkdtempSync(join(tmpdir(), "pi-session-id-readonly-"));
+	tempDirs.push(dir);
+	return dir;
+}
+
+function hasSessionWithId(root: string, sessionId: string): boolean {
+	if (!existsSync(root)) return false;
+	for (const entry of readdirSync(root, { withFileTypes: true })) {
+		const path = join(root, entry.name);
+		if (entry.isDirectory() && hasSessionWithId(path, sessionId)) return true;
+		if (!entry.isFile() || !entry.name.endsWith(".jsonl")) continue;
+
+		try {
+			const firstLine = readFileSync(path, "utf8").split("\n", 1)[0];
+			const header = JSON.parse(firstLine) as { type?: string; id?: string };
+			if (header.type === "session" && header.id === sessionId) return true;
+		} catch {
+			// Ignore malformed session files.
+		}
+	}
+	return false;
+}
+
+interface CliDirs {
+	agentDir: string;
+	projectDir: string;
+	sessionDir: string;
+}
+
+async function runCli(
+	args: string[] | ((dirs: CliDirs) => string[]),
+	setup?: (dirs: CliDirs) => void,
+): Promise<{ code: number | null; agentDir: string; stderr: string }> {
+	const tempRoot = createTempDir();
+	const dirs: CliDirs = {
+		agentDir: join(tempRoot, "agent"),
+		projectDir: join(tempRoot, "project"),
+		sessionDir: join(tempRoot, "sessions"),
+	};
+	mkdirSync(dirs.agentDir, { recursive: true });
+	mkdirSync(dirs.projectDir, { recursive: true });
+	setup?.(dirs);
+	const resolvedArgs = typeof args === "function" ? args(dirs) : args;
+
+	let stderr = "";
+	const code = await new Promise<number | null>((resolvePromise, reject) => {
+		const child = spawn(process.execPath, [cliPath, ...resolvedArgs], {
+			cwd: dirs.projectDir,
+			env: {
+				...process.env,
+				[ENV_AGENT_DIR]: dirs.agentDir,
+				PI_OFFLINE: "1",
+				TSX_TSCONFIG_PATH: resolve(__dirname, "../../../tsconfig.json"),
+			},
+			stdio: ["ignore", "ignore", "pipe"],
+		});
+		child.stderr.on("data", (chunk) => {
+			stderr += chunk.toString();
+		});
+		child.on("error", reject);
+		child.on("close", resolvePromise);
+	});
+
+	return { code, agentDir: dirs.agentDir, stderr };
+}
+
+function writeSession(sessionDir: string, cwd: string, id: string): void {
+	writeFileSync(
+		join(sessionDir, `${id}.jsonl`),
+		`${JSON.stringify({ type: "session", version: 3, id, timestamp: new Date().toISOString(), cwd })}\n`,
+	);
+}
+
+describe("--session-id read-only commands", () => {
+	it("does not reserve a session for --help", async () => {
+		const result = await runCli(["--session-id", "read-only-help", "--help"]);
+
+		expect(result.code).toBe(0);
+		expect(hasSessionWithId(join(result.agentDir, "sessions"), "read-only-help")).toBe(false);
+	});
+
+	it("does not reserve a session for --list-models", async () => {
+		const result = await runCli(["--session-id", "read-only-models", "--list-models"]);
+
+		expect(result.code).toBe(0);
+		expect(hasSessionWithId(join(result.agentDir, "sessions"), "read-only-models")).toBe(false);
+	});
+
+	it("rejects an existing fork target session id", async () => {
+		const result = await runCli(
+			(dirs) => ["--session-dir", dirs.sessionDir, "--fork", "source-id", "--session-id", "existing-id", "-p", "hi"],
+			(dirs) => {
+				mkdirSync(dirs.sessionDir, { recursive: true });
+				writeSession(dirs.sessionDir, dirs.projectDir, "source-id");
+				writeSession(dirs.sessionDir, dirs.projectDir, "existing-id");
+			},
+		);
+
+		expect(result.code).toBe(1);
+		expect(result.stderr).toContain("Session already exists with id 'existing-id'");
+	});
+});
+
+describe("--session-id validation", () => {
+	it("rejects ids invalid under SessionManager rules without stack traces", async () => {
+		for (const id of ["-bad", "bad id"]) {
+			const result = await runCli(["--session-id", id, "-p", "hi"]);
+
+			expect(result.code).toBe(1);
+			expect(result.stderr).toContain("Session id must be non-empty");
+			expect(result.stderr).not.toContain("SessionManager.create");
+		}
+	});
+});

packages/coding-agent/test/session-manager/custom-session-id.test.ts

@@ -1,6 +1,6 @@
-import { mkdtempSync, writeFileSync } from "node:fs";
+import { existsSync, mkdtempSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
-import { join } from "node:path";
+import { basename, join } from "node:path";
 import { describe, expect, it } from "vitest";
 import { SessionManager } from "../../src/core/session-manager.ts";
 
@@ -13,6 +13,23 @@ describe("SessionManager.newSession with custom id", () => {
 		expect(session.getSessionId()).toBe("my-custom-id");
 	});
 
+	it("allows alphanumeric session ids with interior punctuation", () => {
+		const session = SessionManager.inMemory();
+		session.newSession({ id: "abc-123_def.456" });
+		expect(session.getSessionId()).toBe("abc-123_def.456");
+	});
+
+	it("rejects invalid custom session ids", () => {
+		const invalidIds = ["", "-abc", "abc-", "_abc", "abc_", ".abc", "abc.", "abc/def", "abc\\def", "abc def"];
+
+		for (const id of invalidIds) {
+			const session = SessionManager.inMemory();
+			expect(() => session.newSession({ id })).toThrow(
+				"Session id must be non-empty, contain only alphanumeric characters",
+			);
+		}
+	});
+
 	it("generates a UUIDv7 id when no id is provided", () => {
 		const session = SessionManager.inMemory();
 		session.newSession();
@@ -46,6 +63,18 @@ describe("SessionManager.newSession with custom id", () => {
 		expect(session.getHeader()!.id).toBe(session.getSessionId());
 	});
 
+	it("uses the provided id when creating a persisted session", () => {
+		const tempDir = mkdtempSync(join(tmpdir(), "pi-session-manager-"));
+		const session = SessionManager.create(tempDir, tempDir, { id: "created-session-id" });
+
+		expect(session.getSessionId()).toBe("created-session-id");
+		expect(session.getHeader()!.id).toBe("created-session-id");
+		const sessionFile = session.getSessionFile()!;
+		expect(sessionFile).toContain("created-session-id");
+		expect(basename(sessionFile)).toMatch(/^\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}-\d{3}Z_created-session-id\.jsonl$/);
+		expect(existsSync(sessionFile)).toBe(false);
+	});
+
 	it("generates a UUIDv7 id when creating a branched session", () => {
 		const session = SessionManager.inMemory();
 		const firstId = session.appendMessage({
@@ -106,4 +135,28 @@ describe("SessionManager.newSession with custom id", () => {
 		expect(header!.id).toMatch(UUID_V7_RE);
 		expect(header!.parentSession).toBe(sourcePath);
 	});
+
+	it("uses the provided id when forking from another session file", () => {
+		const tempDir = mkdtempSync(join(tmpdir(), "pi-session-manager-"));
+		const sourcePath = join(tempDir, "source.jsonl");
+		writeFileSync(
+			sourcePath,
+			`${JSON.stringify({
+				type: "session",
+				version: 3,
+				id: "source-session-id",
+				timestamp: new Date().toISOString(),
+				cwd: tempDir,
+			})}\n`,
+		);
+
+		const forked = SessionManager.forkFrom(sourcePath, tempDir, tempDir, { id: "forked-session-id" });
+		const header = forked.getHeader();
+		expect(header).not.toBeNull();
+		expect(header!.id).toBe("forked-session-id");
+		expect(header!.parentSession).toBe(sourcePath);
+		const sessionFile = forked.getSessionFile()!;
+		expect(sessionFile).toContain("forked-session-id");
+		expect(basename(sessionFile)).toMatch(/^\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}-\d{3}Z_forked-session-id\.jsonl$/);
+	});
 });

packages/coding-agent/test/session-manager/file-operations.test.ts

@@ -1,4 +1,5 @@
-import { mkdirSync, readFileSync, rmSync, writeFileSync } from "fs";
+import { constants as bufferConstants } from "buffer";
+import { appendFileSync, closeSync, mkdirSync, openSync, readFileSync, rmSync, writeFileSync, writeSync } from "fs";
 import { tmpdir } from "os";
 import { join } from "path";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
@@ -63,6 +64,35 @@ describe("loadEntriesFromFile", () => {
 		const entries = loadEntriesFromFile(file);
 		expect(entries).toHaveLength(2);
 	});
+
+	it("opens session files larger than Node's max string length", () => {
+		const file = join(tempDir, "large.jsonl");
+		writeFileSync(
+			file,
+			'{"type":"session","version":3,"id":"abc","timestamp":"2025-01-01T00:00:00Z","cwd":"/tmp"}\n',
+		);
+
+		const fd = openSync(file, "r+");
+		try {
+			const newline = Buffer.from("\n");
+			const stride = 16 * 1024 * 1024;
+			for (let offset = stride; offset <= bufferConstants.MAX_STRING_LENGTH + stride; offset += stride) {
+				writeSync(fd, newline, 0, newline.length, offset);
+			}
+		} finally {
+			closeSync(fd);
+		}
+
+		appendFileSync(
+			file,
+			'{"type":"message","id":"1","parentId":null,"timestamp":"2025-01-01T00:00:01Z","message":{"role":"user","content":"hi","timestamp":1}}\n',
+		);
+
+		const sessionManager = SessionManager.open(file, tempDir);
+		expect(sessionManager.getSessionId()).toBe("abc");
+		expect(sessionManager.getEntries()).toHaveLength(1);
+		expect(sessionManager.buildSessionContext().messages).toEqual([{ role: "user", content: "hi", timestamp: 1 }]);
+	});
 });
 
 describe("findMostRecentSession", () => {
@@ -124,6 +154,86 @@ describe("findMostRecentSession", () => {
 
 		expect(findMostRecentSession(tempDir)).toBe(valid);
 	});
+
+	it("filters most recent session by cwd", async () => {
+		const projectA = join(tempDir, "project-a");
+		const projectB = join(tempDir, "project-b");
+		const fileA = join(tempDir, "a.jsonl");
+		const fileB = join(tempDir, "b.jsonl");
+
+		writeFileSync(
+			fileA,
+			`${JSON.stringify({ type: "session", id: "a", timestamp: "2025-01-01T00:00:00Z", cwd: projectA })}\n`,
+		);
+		await new Promise((r) => setTimeout(r, 10));
+		writeFileSync(
+			fileB,
+			`${JSON.stringify({ type: "session", id: "b", timestamp: "2025-01-01T00:00:00Z", cwd: projectB })}\n`,
+		);
+
+		expect(findMostRecentSession(tempDir, projectA)).toBe(fileA);
+		expect(findMostRecentSession(tempDir, projectB)).toBe(fileB);
+	});
+});
+
+describe("SessionManager custom flat session directory", () => {
+	let tempDir: string;
+	let projectA: string;
+	let projectB: string;
+
+	beforeEach(() => {
+		tempDir = join(tmpdir(), `session-test-${Date.now()}`);
+		projectA = join(tempDir, "project-a");
+		projectB = join(tempDir, "project-b");
+		mkdirSync(projectA, { recursive: true });
+		mkdirSync(projectB, { recursive: true });
+	});
+
+	afterEach(() => {
+		rmSync(tempDir, { recursive: true, force: true });
+	});
+
+	function createPersistedSession(cwd: string, label: string): string {
+		const session = SessionManager.create(cwd, tempDir);
+		session.appendMessage({ role: "user", content: label, timestamp: Date.now() });
+		session.appendMessage({
+			role: "assistant",
+			content: [{ type: "text", text: `reply to ${label}` }],
+			api: "anthropic-messages",
+			provider: "anthropic",
+			model: "test",
+			usage: {
+				input: 1,
+				output: 1,
+				cacheRead: 0,
+				cacheWrite: 0,
+				totalTokens: 2,
+				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+			},
+			stopReason: "stop",
+			timestamp: Date.now(),
+		});
+		const sessionFile = session.getSessionFile();
+		if (!sessionFile) {
+			throw new Error("Expected persisted session file");
+		}
+		return sessionFile;
+	}
+
+	it("scopes current-folder APIs by cwd while listing all flat sessions", async () => {
+		const sessionA = createPersistedSession(projectA, "from A");
+		await new Promise((r) => setTimeout(r, 10));
+		const sessionB = createPersistedSession(projectB, "from B");
+
+		const currentA = await SessionManager.list(projectA, tempDir);
+		expect(currentA.map((session) => session.path)).toEqual([sessionA]);
+
+		const all = await SessionManager.listAll(tempDir);
+		expect(new Set(all.map((session) => session.path))).toEqual(new Set([sessionA, sessionB]));
+
+		const continuedA = SessionManager.continueRecent(projectA, tempDir);
+		expect(continuedA.getSessionFile()).toBe(sessionA);
+	});
 });
 
 describe("SessionManager.setSessionFile with corrupted files", () => {

packages/coding-agent/test/session-manager/labels.test.ts

@@ -142,6 +142,19 @@ describe("SessionManager labels", () => {
 		expect(msg2Node?.labelTimestamp).toBe(msg2LabelEntry.timestamp);
 	});
 
+	it("rewires children of removed labels when forking", () => {
+		const session = SessionManager.inMemory();
+
+		const msg1Id = session.appendMessage({ role: "user", content: "hello", timestamp: 1 });
+		session.appendLabelChange(msg1Id, "checkpoint");
+		const modelChangeId = session.appendModelChange("anthropic", "claude-test");
+		const msg2Id = session.appendMessage({ role: "user", content: "followup", timestamp: 2 });
+
+		session.createBranchedSession(msg2Id);
+
+		expect(session.getEntry(modelChangeId)?.parentId).toBe(msg1Id);
+	});
+
 	it("labels not on path are not preserved in createBranchedSession", () => {
 		const session = SessionManager.inMemory();
 

packages/coding-agent/test/settings-manager.test.ts

@@ -198,6 +198,24 @@ describe("SettingsManager", () => {
 		});
 	});
 
+	describe("theme setting", () => {
+		it("stores slash-separated automatic theme settings separately from fixed theme names", async () => {
+			const settingsPath = join(agentDir, "settings.json");
+			writeFileSync(settingsPath, JSON.stringify({ theme: "light/dark" }));
+
+			const manager = SettingsManager.create(projectDir, agentDir);
+
+			expect(manager.getTheme()).toBeUndefined();
+			expect(manager.getThemeSetting()).toBe("light/dark");
+
+			manager.setTheme("solarized-light/tokyo-night");
+			await manager.flush();
+
+			const savedSettings = JSON.parse(readFileSync(settingsPath, "utf-8"));
+			expect(savedSettings.theme).toBe("solarized-light/tokyo-night");
+		});
+	});
+
 	describe("error tracking", () => {
 		it("should collect and clear load errors via drainErrors", () => {
 			const globalSettingsPath = join(agentDir, "settings.json");
@@ -214,6 +232,61 @@ describe("SettingsManager", () => {
 		});
 	});
 
+	describe("project trust", () => {
+		it("should skip project settings when project is not trusted", () => {
+			writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ theme: "global" }));
+			writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ theme: "project" }));
+
+			const manager = SettingsManager.create(projectDir, agentDir, { projectTrusted: false });
+
+			expect(manager.isProjectTrusted()).toBe(false);
+			expect(manager.getTheme()).toBe("global");
+			expect(manager.getProjectSettings()).toEqual({});
+		});
+
+		it("should reload project settings after trust changes to true", () => {
+			writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ theme: "global" }));
+			writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ theme: "project" }));
+			const manager = SettingsManager.create(projectDir, agentDir, { projectTrusted: false });
+
+			manager.setProjectTrusted(true);
+
+			expect(manager.isProjectTrusted()).toBe(true);
+			expect(manager.getTheme()).toBe("project");
+		});
+
+		it("should fail project settings writes when project is not trusted", async () => {
+			const projectSettingsPath = join(projectDir, ".pi", "settings.json");
+			writeFileSync(projectSettingsPath, JSON.stringify({ packages: ["npm:existing"] }));
+			const manager = SettingsManager.create(projectDir, agentDir, { projectTrusted: false });
+
+			expect(() => manager.setProjectPackages(["npm:new"])).toThrow(
+				"Project is not trusted; refusing to write project settings",
+			);
+			await manager.flush();
+
+			expect(manager.getProjectSettings()).toEqual({});
+			expect(JSON.parse(readFileSync(projectSettingsPath, "utf-8"))).toEqual({ packages: ["npm:existing"] });
+		});
+
+		it("should read default project trust from global settings only", () => {
+			writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ defaultProjectTrust: "always" }));
+			writeFileSync(join(projectDir, ".pi", "settings.json"), JSON.stringify({ defaultProjectTrust: "never" }));
+
+			const manager = SettingsManager.create(projectDir, agentDir);
+
+			expect(manager.getDefaultProjectTrust()).toBe("always");
+		});
+
+		it("should default invalid project trust settings to ask", () => {
+			writeFileSync(join(agentDir, "settings.json"), JSON.stringify({ defaultProjectTrust: "sometimes" }));
+
+			const manager = SettingsManager.create(projectDir, agentDir);
+
+			expect(manager.getDefaultProjectTrust()).toBe("ask");
+		});
+	});
+
 	describe("project settings directory creation", () => {
 		it("should not create .pi folder when only reading project settings", () => {
 			// Create agent dir with global settings, but NO .pi folder in project

packages/coding-agent/test/startup-session-name.test.ts

@@ -0,0 +1,135 @@
+import { spawn } from "node:child_process";
+import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join, resolve } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { ENV_AGENT_DIR } from "../src/config.ts";
+
+const cliPath = resolve(__dirname, "../src/cli.ts");
+const tempDirs: string[] = [];
+
+afterEach(() => {
+	for (const dir of tempDirs.splice(0)) {
+		rmSync(dir, { recursive: true, force: true });
+	}
+});
+
+function createTempDir(): string {
+	const dir = mkdtempSync(join(tmpdir(), "pi-startup-session-name-"));
+	tempDirs.push(dir);
+	return dir;
+}
+
+interface CliDirs {
+	agentDir: string;
+	projectDir: string;
+	sessionFile: string;
+}
+
+interface CliResult {
+	code: number | null;
+	signal: NodeJS.Signals | null;
+	stderr: string;
+}
+
+function createSessionFile(projectDir: string, sessionFile: string): void {
+	const timestamp = new Date().toISOString();
+	writeFileSync(
+		sessionFile,
+		`${JSON.stringify({ type: "session", version: 3, id: "existing-session", timestamp, cwd: projectDir })}\n${JSON.stringify(
+			{
+				type: "message",
+				id: "assistant-1",
+				parentId: null,
+				timestamp,
+				message: {
+					role: "assistant",
+					content: [{ type: "text", text: "hello" }],
+					provider: "anthropic",
+					model: "claude-sonnet-4-5",
+					timestamp: Date.now(),
+				},
+			},
+		)}\n`,
+	);
+}
+
+function readSessionInfoNames(sessionFile: string): string[] {
+	return readFileSync(sessionFile, "utf8")
+		.trim()
+		.split("\n")
+		.map((line) => JSON.parse(line) as { type?: string; name?: string })
+		.filter((entry) => entry.type === "session_info")
+		.map((entry) => entry.name ?? "");
+}
+
+async function runCli(args: string[], dirs: CliDirs): Promise<CliResult> {
+	let stderr = "";
+	const child = spawn(process.execPath, [cliPath, ...args], {
+		cwd: dirs.projectDir,
+		env: {
+			...process.env,
+			[ENV_AGENT_DIR]: dirs.agentDir,
+			PI_OFFLINE: "1",
+			TSX_TSCONFIG_PATH: resolve(__dirname, "../../../tsconfig.json"),
+		},
+		stdio: ["ignore", "ignore", "pipe"],
+	});
+	child.stderr.on("data", (chunk) => {
+		stderr += chunk.toString();
+	});
+
+	return new Promise((resolvePromise, reject) => {
+		const timeout = setTimeout(() => {
+			child.kill("SIGKILL");
+		}, 10_000);
+		child.on("error", (error) => {
+			clearTimeout(timeout);
+			reject(error);
+		});
+		child.on("close", (code, signal) => {
+			clearTimeout(timeout);
+			resolvePromise({ code, signal, stderr });
+		});
+	});
+}
+
+function setup(): CliDirs {
+	const tempRoot = createTempDir();
+	const dirs = {
+		agentDir: join(tempRoot, "agent"),
+		projectDir: join(tempRoot, "project"),
+		sessionFile: join(tempRoot, "session.jsonl"),
+	};
+	mkdirSync(dirs.agentDir, { recursive: true });
+	mkdirSync(dirs.projectDir, { recursive: true });
+	createSessionFile(dirs.projectDir, dirs.sessionFile);
+	return dirs;
+}
+
+describe("startup session name", () => {
+	it("sets --name on the selected session before runtime model validation", async () => {
+		const dirs = setup();
+		const result = await runCli(
+			["--session", dirs.sessionFile, "--name", "  CLI Named Session  ", "--model", "missing-model", "-p", "hi"],
+			dirs,
+		);
+
+		expect(result.code).toBe(1);
+		expect(result.signal).toBeNull();
+		expect(readSessionInfoNames(dirs.sessionFile)).toEqual(["CLI Named Session"]);
+	});
+
+	it("rejects empty --name values without appending session metadata", async () => {
+		const dirs = setup();
+		const result = await runCli(
+			["--session", dirs.sessionFile, "--name", "   ", "--model", "missing-model", "-p", "hi"],
+			dirs,
+		);
+
+		expect(result.code).toBe(1);
+		expect(result.signal).toBeNull();
+		expect(result.stderr).toContain("--name requires a non-empty value");
+		expect(readSessionInfoNames(dirs.sessionFile)).toEqual([]);
+	});
+});

packages/coding-agent/test/stdout-cleanliness.test.ts

@@ -80,8 +80,24 @@ async function runCli(args: string[]): Promise<{ stdout: string; stderr: string;
 }
 
 describe("stdout cleanliness in non-interactive modes", () => {
-	it("keeps stdout empty for --mode json --help while routing startup chatter to stderr", async () => {
-		const result = await runCli(["--mode", "json", "--help"]);
+	it("prints --version to stdout when stdout is redirected", async () => {
+		const result = await runCli(["--version"]);
+
+		expect(result.code).toBe(0);
+		expect(result.stdout.trim()).toMatch(/^\d+\.\d+\.\d+/);
+		expect(result.stderr).toBe("");
+	});
+
+	it("prints plain --help to stdout when stdout is redirected", async () => {
+		const result = await runCli(["--help"]);
+
+		expect(result.code).toBe(0);
+		expect(result.stdout).toContain("Usage:");
+		expect(result.stderr).not.toContain("Usage:");
+	});
+
+	it("keeps stdout empty for --mode json --help while routing trusted startup chatter to stderr", async () => {
+		const result = await runCli(["--mode", "json", "--help", "--approve"]);
 
 		expect(result.code).toBe(0);
 		expect(result.stdout).toBe("");
@@ -90,13 +106,23 @@ describe("stdout cleanliness in non-interactive modes", () => {
 		expect(result.stderr).toContain("Usage:");
 	});
 
-	it("keeps stdout empty for -p --help while routing startup chatter to stderr", async () => {
+	it("keeps stdout empty for -p --help while routing trusted startup chatter to stderr", async () => {
+		const result = await runCli(["-p", "--help", "--approve"]);
+
+		expect(result.code).toBe(0);
+		expect(result.stdout).toBe("");
+		expect(result.stderr).toContain("changed 1 package in 471ms");
+		expect(result.stderr).toContain("found 0 vulnerabilities");
+		expect(result.stderr).toContain("Usage:");
+	});
+
+	it("ignores untrusted project package installs for help", async () => {
 		const result = await runCli(["-p", "--help"]);
 
 		expect(result.code).toBe(0);
 		expect(result.stdout).toBe("");
-		expect(result.stderr).toContain("changed 1 package in 471ms");
-		expect(result.stderr).toContain("found 0 vulnerabilities");
+		expect(result.stderr).not.toContain("changed 1 package in 471ms");
+		expect(result.stderr).not.toContain("found 0 vulnerabilities");
 		expect(result.stderr).toContain("Usage:");
 	});
 });

packages/coding-agent/test/suite/agent-session-compaction.test.ts

@@ -5,6 +5,7 @@ import {
 	type Model,
 } from "@earendil-works/pi-ai";
 import { afterEach, describe, expect, it, vi } from "vitest";
+import { estimateTokens } from "../../src/core/compaction/index.ts";
 import { createHarness, type Harness } from "./harness.ts";
 
 type SessionWithCompactionInternals = {
@@ -67,19 +68,20 @@ function useSummaryStreamFn(harness: Harness, summary: string): () => number {
 }
 
 function seedCompactableSession(harness: Harness): void {
+	harness.settingsManager.applyOverrides({ compaction: { keepRecentTokens: 1 } });
 	const now = Date.now();
 	harness.sessionManager.appendMessage({
 		role: "user",
 		content: [{ type: "text", text: "message to compact" }],
 		timestamp: now - 1000,
 	});
-	harness.sessionManager.appendMessage(
-		createAssistant(harness, {
-			stopReason: "stop",
-			totalTokens: 100,
-			timestamp: now - 500,
-		}),
-	);
+	const assistant = createAssistant(harness, {
+		stopReason: "stop",
+		totalTokens: 100,
+		timestamp: now - 500,
+	});
+	assistant.content = [{ type: "text", text: "assistant response to compact" }];
+	harness.sessionManager.appendMessage(assistant);
 	harness.session.agent.state.messages = harness.sessionManager.buildSessionContext().messages;
 }
 
@@ -96,6 +98,7 @@ describe("AgentSession compaction characterization", () => {
 
 	it("manually compacts using an extension-provided summary", async () => {
 		const harness = await createHarness({
+			settings: { compaction: { keepRecentTokens: 1 } },
 			extensionFactories: [
 				(pi) => {
 					pi.on("session_before_compact", async (event) => ({
@@ -116,8 +119,10 @@ describe("AgentSession compaction characterization", () => {
 
 		const result = await harness.session.compact();
 		const compactionEntries = harness.sessionManager.getEntries().filter((entry) => entry.type === "compaction");
+		const estimatedTokensAfter = harness.session.messages.reduce((sum, message) => sum + estimateTokens(message), 0);
 
 		expect(result.summary).toBe("summary from extension");
+		expect(result.estimatedTokensAfter).toBe(estimatedTokensAfter);
 		expect(compactionEntries).toHaveLength(1);
 		expect(harness.session.messages[0]?.role).toBe("compactionSummary");
 	});
@@ -145,7 +150,7 @@ describe("AgentSession compaction characterization", () => {
 
 		const result = await harness.session.compact();
 
-		expect(result.summary).toBe("summary from custom stream");
+		expect(result.summary).toContain("summary from custom stream");
 		expect(getStreamCallCount()).toBe(1);
 	});
 
@@ -159,12 +164,15 @@ describe("AgentSession compaction characterization", () => {
 		await sessionInternals._runAutoCompaction("threshold", false);
 
 		const compactionEntries = harness.sessionManager.getEntries().filter((entry) => entry.type === "compaction");
+		const compactionEnd = harness.eventsOfType("compaction_end").at(-1);
 		expect(compactionEntries).toHaveLength(1);
+		expect(compactionEnd?.result?.estimatedTokensAfter).toBeGreaterThan(0);
 		expect(getStreamCallCount()).toBe(1);
 	});
 
 	it("cancels in-progress manual compaction when abortCompaction is called", async () => {
 		const harness = await createHarness({
+			settings: { compaction: { keepRecentTokens: 1 } },
 			extensionFactories: [
 				(pi) => {
 					pi.on("session_before_compact", async (event) => {
@@ -248,6 +256,37 @@ describe("AgentSession compaction characterization", () => {
 		);
 	});
 
+	it("compacts successful overflow responses without retrying", async () => {
+		const harness = await createHarness({
+			settings: { compaction: { enabled: true, keepRecentTokens: 1, reserveTokens: 0 } },
+			models: [{ id: "faux-1", contextWindow: 1, maxTokens: 100 }],
+			extensionFactories: [
+				(pi) => {
+					pi.on("session_before_compact", async (event) => ({
+						compaction: {
+							summary: "successful overflow compacted",
+							firstKeptEntryId: event.preparation.firstKeptEntryId,
+							tokensBefore: event.preparation.tokensBefore,
+							details: {},
+						},
+					}));
+				},
+			],
+		});
+		harnesses.push(harness);
+		harness.setResponses([fauxAssistantMessage("completed answer")]);
+
+		await expect(harness.session.prompt("hello")).resolves.toBeUndefined();
+
+		const compactionEnd = harness.eventsOfType("compaction_end").at(-1);
+		expect(compactionEnd).toMatchObject({
+			reason: "overflow",
+			aborted: false,
+			willRetry: false,
+		});
+		expect(harness.faux.state.callCount).toBe(1);
+	});
+
 	it("ignores stale pre-compaction assistant usage on pre-prompt checks", async () => {
 		const harness = await createHarness();
 		harnesses.push(harness);

packages/coding-agent/test/suite/agent-session-model-extension.test.ts

@@ -2,7 +2,7 @@ import type { AgentTool, ThinkingLevel } from "@earendil-works/pi-agent-core";
 import { fauxAssistantMessage, fauxToolCall, type Model } from "@earendil-works/pi-ai";
 import { Type } from "typebox";
 import { afterEach, describe, expect, it } from "vitest";
-import type { ExtensionAPI } from "../../src/index.ts";
+import type { BuildSystemPromptOptions, ExtensionAPI } from "../../src/index.ts";
 import { createHarness, getAssistantTexts, type Harness } from "./harness.ts";
 
 describe("AgentSession model and extension characterization", () => {
@@ -260,6 +260,34 @@ describe("AgentSession model and extension characterization", () => {
 		expect(extensionApi).toBeDefined();
 	});
 
+	it("allows extension commands to inspect live system prompt options", async () => {
+		const seenOptions: BuildSystemPromptOptions[] = [];
+		const harness = await createHarness({
+			extensionFactories: [
+				(pi) => {
+					pi.registerCommand("inspect-options", {
+						description: "Inspect system prompt options",
+						handler: async (_args, ctx) => {
+							const options = ctx.getSystemPromptOptions();
+							seenOptions.push(options);
+							options.selectedTools?.push("mutated_tool");
+						},
+					});
+				},
+			],
+		});
+		harnesses.push(harness);
+
+		await harness.session.prompt("/inspect-options");
+		await harness.session.prompt("/inspect-options");
+
+		expect(seenOptions).toHaveLength(2);
+		expect(seenOptions[0]).toBe(seenOptions[1]);
+		expect(seenOptions[0]?.cwd).toBe(harness.tempDir);
+		expect(seenOptions[0]?.selectedTools).toContain("read");
+		expect(seenOptions[1]?.selectedTools).toContain("mutated_tool");
+	});
+
 	it("allows before_agent_start handlers to inject custom messages and modify the system prompt", async () => {
 		const harness = await createHarness({
 			extensionFactories: [

packages/coding-agent/test/suite/agent-session-prompt.test.ts

@@ -5,6 +5,7 @@ import type { AgentTool } from "@earendil-works/pi-agent-core";
 import { fauxAssistantMessage, fauxToolCall, type Model } from "@earendil-works/pi-ai";
 import { Type } from "typebox";
 import { afterEach, describe, expect, it } from "vitest";
+import type { InputEvent } from "../../src/core/extensions/index.ts";
 import type { PromptTemplate } from "../../src/core/prompt-templates.ts";
 import { createSyntheticSourceInfo } from "../../src/core/source-info.ts";
 import { createTestResourceLoader } from "../utilities.ts";
@@ -259,6 +260,80 @@ describe("AgentSession prompt characterization", () => {
 		expect(getMessageText(harness.session.messages[0]!)).toBe("from extension");
 	});
 
+	it("does not report streamingBehavior to input handlers while idle", async () => {
+		const inputEvents: InputEvent[] = [];
+		const harness = await createHarness({
+			extensionFactories: [
+				(pi) => {
+					pi.on("input", (event) => {
+						inputEvents.push(event);
+					});
+				},
+			],
+		});
+		harnesses.push(harness);
+		harness.setResponses([fauxAssistantMessage("ok")]);
+
+		await harness.session.prompt("idle", { streamingBehavior: "followUp" });
+
+		expect(inputEvents).toHaveLength(1);
+		expect(inputEvents[0]?.streamingBehavior).toBeUndefined();
+	});
+
+	it("reports streamingBehavior to input handlers while streaming", async () => {
+		let releaseToolExecution: (() => void) | undefined;
+		const toolRelease = new Promise<void>((resolve) => {
+			releaseToolExecution = resolve;
+		});
+		const inputEvents: InputEvent[] = [];
+		const waitTool: AgentTool = {
+			name: "wait",
+			label: "Wait",
+			description: "Wait for release",
+			parameters: Type.Object({}),
+			execute: async () => {
+				await toolRelease;
+				return {
+					content: [{ type: "text", text: "released" }],
+					details: {},
+				};
+			},
+		};
+		const harness = await createHarness({
+			tools: [waitTool],
+			extensionFactories: [
+				(pi) => {
+					pi.on("input", (event) => {
+						inputEvents.push(event);
+					});
+				},
+			],
+		});
+		harnesses.push(harness);
+		harness.setResponses([
+			fauxAssistantMessage(fauxToolCall("wait", {}), { stopReason: "toolUse" }),
+			fauxAssistantMessage("done"),
+		]);
+
+		const sawToolStart = new Promise<void>((resolve) => {
+			const unsubscribe = harness.session.subscribe((event) => {
+				if (event.type === "tool_execution_start") {
+					unsubscribe();
+					resolve();
+				}
+			});
+		});
+
+		const promptPromise = harness.session.prompt("start");
+		await sawToolStart;
+		await harness.session.prompt("queued", { streamingBehavior: "followUp" });
+
+		expect(inputEvents.map((event) => event.streamingBehavior)).toEqual([undefined, "followUp"]);
+
+		releaseToolExecution?.();
+		await promptPromise;
+	});
+
 	it("throws when prompted during streaming without a streamingBehavior", async () => {
 		let releaseToolExecution: (() => void) | undefined;
 		const toolRelease = new Promise<void>((resolve) => {

packages/coding-agent/test/suite/agent-session-queue.test.ts

@@ -419,4 +419,27 @@ describe("AgentSession queue characterization", () => {
 			'Extension command "/testcmd" cannot be queued. Use prompt() or execute the command when not streaming.',
 		);
 	});
+
+	it("delivers follow-ups queued during agent_end", async () => {
+		let sent = false;
+		const harness = await createHarness({
+			extensionFactories: [
+				(pi: ExtensionAPI) => {
+					pi.on("agent_end", async () => {
+						if (sent) return;
+						sent = true;
+						pi.sendUserMessage("conflict report", { deliverAs: "followUp" });
+					});
+				},
+			],
+		});
+		harnesses.push(harness);
+
+		harness.setResponses([fauxAssistantMessage("reply"), fauxAssistantMessage("follow-up reply")]);
+
+		await harness.session.prompt("hello");
+		await harness.session.agent.waitForIdle();
+
+		expect(getUserTexts(harness)).toEqual(["hello", "conflict report"]);
+	});
 });

packages/coding-agent/test/suite/harness.ts

@@ -60,6 +60,9 @@ export interface HarnessOptions {
 	settings?: Partial<Settings>;
 	systemPrompt?: string;
 	tools?: AgentTool[];
+	initialActiveToolNames?: string[];
+	allowedToolNames?: string[];
+	excludedToolNames?: string[];
 	resourceLoader?: ResourceLoader;
 	extensionFactories?: Array<ExtensionFactory | CreateTestExtensionsResultInput>;
 	withConfiguredAuth?: boolean;
@@ -173,6 +176,9 @@ export async function createHarness(options: HarnessOptions = {}): Promise<Harne
 		modelRegistry,
 		resourceLoader,
 		baseToolsOverride: toolMap,
+		initialActiveToolNames: options.initialActiveToolNames,
+		allowedToolNames: options.allowedToolNames,
+		excludedToolNames: options.excludedToolNames,
 		extensionRunnerRef,
 	});
 

packages/coding-agent/test/suite/regressions/5080-signal-shutdown-extension-cleanup.test.ts

@@ -0,0 +1,185 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import chalk from "chalk";
+import { afterEach, describe, expect, test, vi } from "vitest";
+import { APP_NAME } from "../../../src/config.ts";
+import type { SessionManager } from "../../../src/core/session-manager.ts";
+import { InteractiveMode } from "../../../src/modes/interactive/interactive-mode.ts";
+
+// Regression for https://github.com/earendil-works/pi/issues/5080
+//
+// On SIGTERM/SIGHUP the graceful shutdown must emit `session_shutdown`
+// (runtimeHost.dispose) BEFORE touching the terminal. Extension teardown such
+// as removing a socket does not write to the tty, so it must not be skipped if
+// a later terminal-restore write fails on a dead or stalled terminal. The
+// interactive quit path (Ctrl+D, /quit) keeps the opposite order to preserve
+// the final TUI frame.
+
+type ShutdownThis = {
+	isShuttingDown: boolean;
+	unregisterSignalHandlers: () => void;
+	runtimeHost: { dispose: () => Promise<void> };
+	ui: { terminal: { drainInput: (ms: number) => Promise<void> } };
+	themeController: { disableAutoSync: () => void };
+	stop: () => void;
+	sessionManager: SessionManager;
+};
+
+type InteractiveModePrototypeWithShutdown = {
+	shutdown(this: ShutdownThis, options?: { fromSignal?: boolean }): Promise<void>;
+};
+
+const interactiveModePrototype = InteractiveMode.prototype as unknown;
+const tempDirs: string[] = [];
+const originalStdoutIsTTY = Object.getOwnPropertyDescriptor(process.stdout, "isTTY");
+
+class ProcessExitError extends Error {}
+
+function createSessionManager(options: { sessionFile?: string } = {}): SessionManager {
+	return {
+		isPersisted: () => options.sessionFile !== undefined,
+		getSessionFile: () => options.sessionFile,
+		getSessionId: () => "test-session",
+		getSessionDir: () => "/tmp/pi-sessions",
+		usesDefaultSessionDir: () => true,
+	} as unknown as SessionManager;
+}
+
+function createTempFile(): string {
+	const dir = mkdtempSync(join(tmpdir(), "pi-shutdown-resume-hint-"));
+	tempDirs.push(dir);
+	const file = join(dir, "session.jsonl");
+	writeFileSync(file, "\n");
+	return file;
+}
+
+function setStdoutIsTTY(value: boolean): void {
+	Object.defineProperty(process.stdout, "isTTY", { configurable: true, value });
+}
+
+function restoreStdoutIsTTY(): void {
+	if (originalStdoutIsTTY) {
+		Object.defineProperty(process.stdout, "isTTY", originalStdoutIsTTY);
+	} else {
+		Reflect.deleteProperty(process.stdout, "isTTY");
+	}
+}
+
+function createContext(order: string[], sessionManager = createSessionManager()): ShutdownThis {
+	return {
+		isShuttingDown: false,
+		unregisterSignalHandlers: vi.fn(),
+		runtimeHost: {
+			dispose: vi.fn(async () => {
+				order.push("dispose");
+			}),
+		},
+		ui: {
+			terminal: {
+				drainInput: vi.fn(async () => {
+					order.push("drainInput");
+				}),
+			},
+		},
+		themeController: { disableAutoSync: vi.fn() },
+		stop: vi.fn(() => {
+			order.push("stop");
+		}),
+		sessionManager,
+	};
+}
+
+async function callShutdown(context: ShutdownThis, options?: { fromSignal?: boolean }): Promise<void> {
+	try {
+		await (interactiveModePrototype as InteractiveModePrototypeWithShutdown).shutdown.call(context, options);
+	} catch (error) {
+		if (!(error instanceof ProcessExitError)) throw error;
+	}
+}
+
+describe("InteractiveMode.shutdown ordering (#5080)", () => {
+	afterEach(() => {
+		vi.restoreAllMocks();
+		restoreStdoutIsTTY();
+		for (const dir of tempDirs.splice(0)) {
+			rmSync(dir, { recursive: true, force: true });
+		}
+	});
+
+	test("signal-triggered shutdown emits session_shutdown before terminal writes", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+		const order: string[] = [];
+		const context = createContext(order);
+
+		await callShutdown(context, { fromSignal: true });
+
+		expect(order).toEqual(["dispose", "drainInput", "stop"]);
+		expect(context.isShuttingDown).toBe(true);
+	});
+
+	test("interactive quit stops the TUI before emitting session_shutdown", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+		const order: string[] = [];
+		const context = createContext(order);
+
+		await callShutdown(context);
+
+		expect(order).toEqual(["drainInput", "stop", "dispose"]);
+	});
+
+	test("interactive quit prints a resume hint for persisted sessions", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+		const stdoutWrite = vi
+			.spyOn(process.stdout, "write")
+			.mockImplementation((() => true) as typeof process.stdout.write);
+		setStdoutIsTTY(true);
+		const order: string[] = [];
+		const context = createContext(order, createSessionManager({ sessionFile: createTempFile() }));
+
+		await callShutdown(context);
+
+		expect(order).toEqual(["drainInput", "stop", "dispose"]);
+		expect(stdoutWrite).toHaveBeenCalledWith(
+			`${chalk.dim("To resume this session:")} ${APP_NAME} --session test-session\n`,
+		);
+	});
+
+	test("signal-triggered shutdown does not print a resume hint", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+		const stdoutWrite = vi
+			.spyOn(process.stdout, "write")
+			.mockImplementation((() => true) as typeof process.stdout.write);
+		setStdoutIsTTY(true);
+		const order: string[] = [];
+		const context = createContext(order, createSessionManager({ sessionFile: createTempFile() }));
+
+		await callShutdown(context, { fromSignal: true });
+
+		for (const call of stdoutWrite.mock.calls) {
+			expect(call[0]).not.toContain("To resume this session:");
+		}
+	});
+
+	test("re-entrant shutdown is a no-op", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+		const order: string[] = [];
+		const context = createContext(order);
+		context.isShuttingDown = true;
+
+		await callShutdown(context, { fromSignal: true });
+
+		expect(order).toEqual([]);
+		expect(context.runtimeHost.dispose).not.toHaveBeenCalled();
+	});
+});

packages/coding-agent/test/suite/regressions/5109-exclude-tools.test.ts

@@ -0,0 +1,81 @@
+import { Type } from "typebox";
+import { describe, expect, it } from "vitest";
+import type { ExtensionFactory } from "../../../src/index.ts";
+import { createHarness } from "../harness.ts";
+
+function toolNames(tools: Array<{ name: string }>): string[] {
+	return tools.map((tool) => tool.name).sort();
+}
+
+describe("regression #5109: exclude tools", () => {
+	const extensionFactories: ExtensionFactory[] = [
+		(pi) => {
+			pi.on("session_start", () => {
+				pi.registerTool({
+					name: "ask_question",
+					label: "Ask Question",
+					description: "Ask a question",
+					promptSnippet: "Ask a question",
+					parameters: Type.Object({}),
+					execute: async () => ({
+						content: [{ type: "text", text: "ok" }],
+						details: {},
+					}),
+				});
+				pi.registerTool({
+					name: "dynamic_tool",
+					label: "Dynamic Tool",
+					description: "Dynamic test tool",
+					promptSnippet: "Run dynamic test behavior",
+					parameters: Type.Object({}),
+					execute: async () => ({
+						content: [{ type: "text", text: "ok" }],
+						details: {},
+					}),
+				});
+			});
+		},
+	];
+
+	it("filters built-in and extension tools from available and active tools", async () => {
+		const harness = await createHarness({
+			excludedToolNames: ["read", "ask_question"],
+			extensionFactories,
+		});
+		try {
+			await harness.session.bindExtensions({});
+
+			const allToolNames = toolNames(harness.session.getAllTools());
+			expect(allToolNames).not.toContain("read");
+			expect(allToolNames).not.toContain("ask_question");
+			expect(allToolNames).toContain("bash");
+			expect(allToolNames).toContain("dynamic_tool");
+			expect(harness.session.getActiveToolNames().sort()).toEqual(["bash", "dynamic_tool", "edit", "write"]);
+			expect(harness.session.systemPrompt).not.toContain("- read:");
+			expect(harness.session.systemPrompt).not.toContain("ask_question");
+			expect(harness.session.systemPrompt).toContain("- dynamic_tool: Run dynamic test behavior");
+		} finally {
+			harness.cleanup();
+		}
+	});
+
+	it("lets excluded tools override the allowlist", async () => {
+		const harness = await createHarness({
+			allowedToolNames: ["read", "bash", "ask_question"],
+			excludedToolNames: ["read", "ask_question"],
+			initialActiveToolNames: ["read", "bash", "ask_question"],
+			extensionFactories,
+		});
+		try {
+			await harness.session.bindExtensions({});
+
+			expect(toolNames(harness.session.getAllTools())).toEqual(["bash"]);
+			expect(harness.session.getActiveToolNames()).toEqual(["bash"]);
+			expect(harness.session.systemPrompt).toContain("- bash:");
+			expect(harness.session.systemPrompt).not.toContain("- read:");
+			expect(harness.session.systemPrompt).not.toContain("ask_question");
+		} finally {
+			harness.cleanup();
+		}
+	});
+});

packages/coding-agent/test/suite/regressions/5208-late-bash-output.test.ts

@@ -0,0 +1,29 @@
+import { describe, expect, it } from "vitest";
+import { type BashOperations, createBashTool } from "../../../src/core/tools/bash.ts";
+
+function getTextOutput(result: { content?: Array<{ type: string; text?: string }> }): string {
+	return (
+		result.content
+			?.filter((block) => block.type === "text")
+			.map((block) => block.text ?? "")
+			.join("\n") ?? ""
+	);
+}
+
+describe("regression #5208: late bash output callbacks", () => {
+	it("ignores output callbacks after bash operations resolve", async () => {
+		const operations: BashOperations = {
+			exec: async (_command, _cwd, { onData }) => {
+				onData(Buffer.from("before\n", "utf-8"));
+				setTimeout(() => onData(Buffer.from("late\n", "utf-8")), 0);
+				return { exitCode: 0 };
+			},
+		};
+		const bash = createBashTool(process.cwd(), { operations });
+
+		const result = await bash.execute("test-call-late-output", { command: "late-output" });
+		await new Promise((resolve) => setTimeout(resolve, 20));
+
+		expect(getTextOutput(result).trim()).toBe("before");
+	});
+});

packages/coding-agent/test/suite/regressions/5303-bash-output-truncation.test.ts

@@ -0,0 +1,79 @@
+import type { ChildProcessByStdio } from "node:child_process";
+import type { Readable } from "node:stream";
+import { afterEach, describe, expect, it } from "vitest";
+import { spawnProcess, waitForChildProcess } from "../../../src/utils/child-process.ts";
+
+/**
+ * Regression test for https://github.com/earendil-works/pi/issues/5303
+ *
+ * waitForChildProcess armed a fixed 100ms timer on `exit` and destroyed the
+ * stdio streams when it fired. When a short-lived detached descendant kept the
+ * stdout pipe open, `close` never fired, so that timer was the only thing that
+ * resolved the wait, and any output written more than 100ms after exit was
+ * binned. In practice every git commit whose pre-commit hook runs lint-staged
+ * came back truncated mid-listr2 output, read by the model as a hang.
+ *
+ * The fix re-arms the grace on each chunk, so an actively writing pipe keeps us
+ * reading while a genuinely idle held-open handle still releases after the
+ * grace elapses. Both behaviours are covered below.
+ */
+describe.skipIf(process.platform === "win32")("issue #5303 bash output truncation past exit", () => {
+	let child: ChildProcessByStdio<null, Readable, Readable> | undefined;
+
+	afterEach(() => {
+		if (child?.pid) {
+			try {
+				process.kill(-child.pid, "SIGKILL");
+			} catch {
+				// Already gone.
+			}
+		}
+		child = undefined;
+	});
+
+	it("captures output emitted after exit while a detached child holds stdout open", async () => {
+		// The shell exits immediately, but a backgrounded subshell keeps the stdout
+		// pipe open and emits ticks every 50ms, the last well past the 100ms grace.
+		const command = 'printf "HEAD\\n"; ( for i in 1 2 3 4 5 6; do sleep 0.05; printf "TICK$i\\n"; done ) &';
+		child = spawnProcess("/bin/sh", ["-c", command], {
+			stdio: ["ignore", "pipe", "pipe"],
+			detached: true,
+		}) as ChildProcessByStdio<null, Readable, Readable>;
+
+		let output = "";
+		child.stdout.on("data", (chunk: Buffer) => {
+			output += chunk.toString();
+		});
+
+		const exitCode = await waitForChildProcess(child);
+
+		expect(exitCode).toBe(0);
+		expect(output).toContain("HEAD");
+		expect(output).toContain("TICK6");
+	});
+
+	it("resolves promptly when a detached child holds stdout open but stays quiet", async () => {
+		// The shell exits, but a backgrounded sleeper inherits the stdout pipe and
+		// keeps it open for a long time without writing. `close` never fires, so we
+		// must still release via the idle grace rather than hang on the open handle.
+		const command = 'printf "DONE\\n"; ( sleep 30 ) &';
+		child = spawnProcess("/bin/sh", ["-c", command], {
+			stdio: ["ignore", "pipe", "pipe"],
+			detached: true,
+		}) as ChildProcessByStdio<null, Readable, Readable>;
+
+		let output = "";
+		child.stdout.on("data", (chunk: Buffer) => {
+			output += chunk.toString();
+		});
+
+		const start = Date.now();
+		const exitCode = await waitForChildProcess(child);
+		const elapsed = Date.now() - start;
+
+		expect(exitCode).toBe(0);
+		expect(output).toContain("DONE");
+		// Must not wait for the 30s sleeper; the idle grace releases us in well under a second.
+		expect(elapsed).toBeLessThan(2000);
+	});
+});

packages/coding-agent/test/suite/regressions/5433-extension-oauth-prompt-input.test.ts

@@ -0,0 +1,93 @@
+import { setKeybindings, type TUI } from "@earendil-works/pi-tui";
+import { beforeAll, beforeEach, describe, expect, test, vi } from "vitest";
+import { KeybindingsManager } from "../../../src/core/keybindings.ts";
+import { LoginDialogComponent } from "../../../src/modes/interactive/components/login-dialog.ts";
+import { initTheme } from "../../../src/modes/interactive/theme/theme.ts";
+import { stripAnsi } from "../../../src/utils/ansi.ts";
+
+vi.mock("../../../src/utils/open-browser.ts", () => ({
+	openBrowser: vi.fn(),
+}));
+
+function createDialog(): LoginDialogComponent {
+	return new LoginDialogComponent(
+		{ requestRender: vi.fn() } as unknown as TUI,
+		"prompt-repro",
+		() => {},
+		"Prompt Repro",
+	);
+}
+
+function renderDialog(dialog: LoginDialogComponent): string[] {
+	return stripAnsi(dialog.render(120).join("\n"))
+		.split("\n")
+		.map((line) => line.trimEnd());
+}
+
+function countRenderedValue(lines: string[], value: string): number {
+	return lines.filter((line) => line.trim() === `> ${value}`).length;
+}
+
+describe("LoginDialogComponent OAuth prompts", () => {
+	beforeAll(() => {
+		initTheme("dark");
+	});
+
+	beforeEach(() => {
+		setKeybindings(new KeybindingsManager());
+	});
+
+	test("keeps previous prompt input stable when a later prompt is active", async () => {
+		const dialog = createDialog();
+
+		const firstPrompt = dialog.showPrompt("First prompt:", "first-value");
+		dialog.handleInput("first-value");
+		dialog.handleInput("\n");
+		await expect(firstPrompt).resolves.toBe("first-value");
+
+		const secondPrompt = dialog.showPrompt("Second prompt:");
+		dialog.handleInput("second-secret-demo");
+
+		const lines = renderDialog(dialog);
+		expect(lines.join("\n")).toContain("First prompt:");
+		expect(lines.join("\n")).toContain("Second prompt:");
+		expect(countRenderedValue(lines, "first-value")).toBe(1);
+		expect(countRenderedValue(lines, "second-secret-demo")).toBe(1);
+
+		dialog.handleInput("\n");
+		await expect(secondPrompt).resolves.toBe("second-secret-demo");
+	});
+
+	test("preserves auth instructions when showing a prompt", () => {
+		const dialog = createDialog();
+
+		dialog.showAuth("https://example.invalid/login", "Authorize the extension");
+		dialog.showPrompt("First prompt:");
+
+		const output = renderDialog(dialog).join("\n");
+		expect(output).toContain("https://example.invalid/login");
+		expect(output).toContain("Authorize the extension");
+		expect(output).toContain("First prompt:");
+	});
+
+	test("keeps previous manual input stable when a later prompt is active", async () => {
+		const dialog = createDialog();
+
+		const manualInput = dialog.showManualInput("Paste callback URL:");
+		dialog.handleInput("callback-value");
+		dialog.handleInput("\n");
+		await expect(manualInput).resolves.toBe("callback-value");
+
+		const prompt = dialog.showPrompt("Second prompt:");
+		dialog.handleInput("second-secret-demo");
+
+		const lines = renderDialog(dialog);
+		expect(lines.join("\n")).toContain("Paste callback URL:");
+		expect(lines.join("\n")).toContain("Second prompt:");
+		expect(countRenderedValue(lines, "callback-value")).toBe(1);
+		expect(countRenderedValue(lines, "second-secret-demo")).toBe(1);
+
+		dialog.handleInput("\n");
+		await expect(prompt).resolves.toBe("second-secret-demo");
+	});
+});

packages/coding-agent/test/suite/regressions/5596-missing-theme-export.test.ts

@@ -0,0 +1,89 @@
+import { existsSync, mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { Agent } from "@earendil-works/pi-agent-core";
+import { fauxAssistantMessage, registerFauxProvider } from "@earendil-works/pi-ai";
+import { afterEach, describe, expect, it } from "vitest";
+import { AgentSession } from "../../../src/core/agent-session.ts";
+import { AuthStorage } from "../../../src/core/auth-storage.ts";
+import { convertToLlm } from "../../../src/core/messages.ts";
+import { ModelRegistry } from "../../../src/core/model-registry.ts";
+import { SessionManager } from "../../../src/core/session-manager.ts";
+import { SettingsManager } from "../../../src/core/settings-manager.ts";
+import { initTheme } from "../../../src/modes/interactive/theme/theme.ts";
+import { createTestResourceLoader } from "../../utilities.ts";
+
+describe("regression #5596: missing configured theme export", () => {
+	const cleanups: Array<() => void> = [];
+
+	afterEach(() => {
+		while (cleanups.length > 0) {
+			cleanups.pop()?.();
+		}
+		initTheme("dark");
+	});
+
+	it("exports with the active fallback theme when the configured theme is missing", async () => {
+		const tempDir = mkdtempSync(join(tmpdir(), "pi-5596-"));
+		const faux = registerFauxProvider({
+			models: [{ id: "faux-1", reasoning: false }],
+		});
+		faux.setResponses([fauxAssistantMessage("hello")]);
+
+		const model = faux.getModel();
+		const authStorage = AuthStorage.inMemory();
+		authStorage.setRuntimeApiKey(model.provider, "faux-key");
+		const modelRegistry = ModelRegistry.inMemory(authStorage);
+		modelRegistry.registerProvider(model.provider, {
+			baseUrl: model.baseUrl,
+			apiKey: "faux-key",
+			api: faux.api,
+			models: faux.models.map((registeredModel) => ({
+				id: registeredModel.id,
+				name: registeredModel.name,
+				api: registeredModel.api,
+				reasoning: registeredModel.reasoning,
+				input: registeredModel.input,
+				cost: registeredModel.cost,
+				contextWindow: registeredModel.contextWindow,
+				maxTokens: registeredModel.maxTokens,
+				baseUrl: registeredModel.baseUrl,
+			})),
+		});
+
+		const settingsManager = SettingsManager.inMemory({ theme: "missing-theme" });
+		const sessionManager = SessionManager.create(tempDir, join(tempDir, "sessions"));
+		const agent = new Agent({
+			getApiKey: () => "faux-key",
+			initialState: {
+				model,
+				systemPrompt: "You are a test assistant.",
+				tools: [],
+			},
+			convertToLlm,
+		});
+		const session = new AgentSession({
+			agent,
+			sessionManager,
+			settingsManager,
+			cwd: tempDir,
+			modelRegistry,
+			resourceLoader: createTestResourceLoader(),
+		});
+		cleanups.push(() => {
+			session.dispose();
+			faux.unregister();
+			if (existsSync(tempDir)) {
+				rmSync(tempDir, { recursive: true, force: true });
+			}
+		});
+
+		await session.prompt("hi");
+		initTheme(settingsManager.getTheme());
+
+		const outputPath = join(tempDir, "export.html");
+		await expect(session.exportToHtml(outputPath)).resolves.toBe(outputPath);
+		expect(existsSync(outputPath)).toBe(true);
+		expect(settingsManager.getTheme()).toBe("missing-theme");
+	});
+});

packages/coding-agent/test/suite/regressions/5661-uppercase-header-values.test.ts

@@ -0,0 +1,91 @@
+import { readFileSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+import { afterEach, describe, expect, it } from "vitest";
+import { ENV_AGENT_DIR } from "../../../src/config.ts";
+import { AuthStorage } from "../../../src/core/auth-storage.ts";
+import { ModelRegistry } from "../../../src/core/model-registry.ts";
+import { runMigrations } from "../../../src/migrations.ts";
+import { createHarness } from "../harness.ts";
+
+describe("regression #5661: uppercase models.json header values", () => {
+	const cleanups: Array<() => void> = [];
+
+	afterEach(() => {
+		while (cleanups.length > 0) {
+			cleanups.pop()?.();
+		}
+	});
+
+	function withAgentDir(agentDir: string, fn: () => void): void {
+		const previousAgentDir = process.env[ENV_AGENT_DIR];
+		process.env[ENV_AGENT_DIR] = agentDir;
+		try {
+			fn();
+		} finally {
+			if (previousAgentDir === undefined) {
+				delete process.env[ENV_AGENT_DIR];
+			} else {
+				process.env[ENV_AGENT_DIR] = previousAgentDir;
+			}
+		}
+	}
+
+	it("keeps uppercase header strings as literals during startup migrations", async () => {
+		const harness = await createHarness({ withConfiguredAuth: false });
+		cleanups.push(harness.cleanup);
+
+		const envKeys = ["CUSTOM_API_KEY", "BEARER"];
+		const savedEnv: Record<string, string | undefined> = {};
+		for (const key of envKeys) {
+			savedEnv[key] = process.env[key];
+			process.env[key] = `env-${key}`;
+		}
+		cleanups.push(() => {
+			for (const key of envKeys) {
+				if (savedEnv[key] === undefined) {
+					delete process.env[key];
+				} else {
+					process.env[key] = savedEnv[key];
+				}
+			}
+		});
+
+		const modelsPath = join(harness.tempDir, "models.json");
+		writeFileSync(
+			modelsPath,
+			`${JSON.stringify(
+				{
+					providers: {
+						"my-provider": {
+							baseUrl: "https://example.com/v1",
+							apiKey: "CUSTOM_API_KEY",
+							api: "openai-completions",
+							headers: { Authorization: "BEARER" },
+							models: [{ id: "my-model" }],
+						},
+					},
+				},
+				null,
+				2,
+			)}\n`,
+			"utf-8",
+		);
+
+		withAgentDir(harness.tempDir, () => runMigrations(harness.tempDir));
+
+		const migrated = JSON.parse(readFileSync(modelsPath, "utf-8")) as {
+			providers: Record<string, { apiKey?: string; headers?: Record<string, string> }>;
+		};
+		expect(migrated.providers["my-provider"]?.apiKey).toBe("CUSTOM_API_KEY");
+		expect(migrated.providers["my-provider"]?.headers?.Authorization).toBe("BEARER");
+
+		const registry = ModelRegistry.create(AuthStorage.create(join(harness.tempDir, "auth.json")), modelsPath);
+		const model = registry.find("my-provider", "my-model");
+		expect(model).toBeDefined();
+		expect(await registry.getApiKeyAndHeaders(model!)).toMatchObject({
+			ok: true,
+			apiKey: "CUSTOM_API_KEY",
+			headers: { Authorization: "BEARER" },
+		});
+	});
+});

packages/coding-agent/test/suite/regressions/5724-sigterm-signal-exit.test.ts

@@ -0,0 +1,94 @@
+import { afterEach, describe, expect, test, vi } from "vitest";
+import { InteractiveMode } from "../../../src/modes/interactive/interactive-mode.ts";
+
+// Regression for https://github.com/earendil-works/pi/issues/5724
+//
+// `proper-lockfile` installs `signal-exit`, whose signal listener re-sends
+// SIGTERM/SIGHUP when it observes no other process listeners during the same
+// signal dispatch. InteractiveMode must therefore keep its signal handlers
+// registered until async terminal cleanup has completed.
+
+type ShutdownThis = {
+	isShuttingDown: boolean;
+	unregisterSignalHandlers: () => void;
+	runtimeHost: { dispose: () => Promise<void> };
+	ui: { terminal: { drainInput: (ms: number) => Promise<void> } };
+	themeController: { disableAutoSync: () => void };
+	stop: () => void;
+};
+
+type InteractiveModePrototypeWithShutdown = {
+	shutdown(this: ShutdownThis, options?: { fromSignal?: boolean }): Promise<void>;
+};
+
+const interactiveModePrototype = InteractiveMode.prototype as unknown;
+
+class ProcessExitError extends Error {}
+
+function deferred(): { promise: Promise<void>; resolve: () => void } {
+	let resolve: (() => void) | undefined;
+	const promise = new Promise<void>((res) => {
+		resolve = res;
+	});
+	return {
+		promise,
+		resolve: () => resolve?.(),
+	};
+}
+
+async function callShutdown(context: ShutdownThis, options?: { fromSignal?: boolean }): Promise<void> {
+	try {
+		await (interactiveModePrototype as InteractiveModePrototypeWithShutdown).shutdown.call(context, options);
+	} catch (error) {
+		if (!(error instanceof ProcessExitError)) throw error;
+	}
+}
+
+describe("InteractiveMode SIGTERM shutdown with signal-exit (#5724)", () => {
+	afterEach(() => {
+		vi.restoreAllMocks();
+	});
+
+	test("keeps signal handlers registered while signal-triggered cleanup is pending", async () => {
+		vi.spyOn(process, "exit").mockImplementation((() => {
+			throw new ProcessExitError();
+		}) as typeof process.exit);
+
+		const order: string[] = [];
+		const dispose = deferred();
+		const context: ShutdownThis = {
+			isShuttingDown: false,
+			unregisterSignalHandlers: vi.fn(() => {
+				order.push("unregister");
+			}),
+			runtimeHost: {
+				dispose: vi.fn(() => {
+					order.push("dispose");
+					return dispose.promise;
+				}),
+			},
+			ui: {
+				terminal: {
+					drainInput: vi.fn(async () => {
+						order.push("drainInput");
+					}),
+				},
+			},
+			themeController: { disableAutoSync: vi.fn() },
+			stop: vi.fn(() => {
+				order.push("stop");
+			}),
+		};
+
+		const shutdownPromise = callShutdown(context, { fromSignal: true });
+		await Promise.resolve();
+
+		expect(order).toEqual(["dispose"]);
+		expect(context.unregisterSignalHandlers).not.toHaveBeenCalled();
+
+		dispose.resolve();
+		await shutdownPromise;
+
+		expect(order).toEqual(["dispose", "drainInput", "stop"]);
+	});
+});

packages/coding-agent/test/suite/regressions/5868-rpc-unknown-command-id.test.ts

@@ -0,0 +1,113 @@
+import { afterEach, describe, expect, test, vi } from "vitest";
+import type { AgentSessionRuntime } from "../../../src/core/agent-session-runtime.ts";
+import { runRpcMode } from "../../../src/modes/rpc/rpc-mode.ts";
+import { createHarness, type Harness } from "../harness.ts";
+
+// Regression for https://github.com/earendil-works/pi/issues/5868
+
+const rpcIo = vi.hoisted(() => ({
+	outputLines: [] as string[],
+	lineHandler: undefined as ((line: string) => void) | undefined,
+}));
+
+vi.mock("../../../src/core/output-guard.js", () => ({
+	flushRawStdout: vi.fn(async () => {}),
+	takeOverStdout: vi.fn(),
+	waitForRawStdoutBackpressure: vi.fn(async () => {}),
+	writeRawStdout: (line: string) => {
+		rpcIo.outputLines.push(line);
+	},
+}));
+
+vi.mock("../../../src/modes/interactive/theme/theme.js", () => ({ theme: {} }));
+
+vi.mock("../../../src/modes/rpc/jsonl.js", () => ({
+	attachJsonlLineReader: vi.fn((_stream: NodeJS.ReadableStream, onLine: (line: string) => void) => {
+		rpcIo.lineHandler = onLine;
+		return () => {
+			rpcIo.lineHandler = undefined;
+		};
+	}),
+	serializeJsonLine: (value: unknown) => `${JSON.stringify(value)}\n`,
+}));
+
+type NodeListener = Parameters<typeof process.on>[1];
+
+type ListenerSnapshot = {
+	stdinEnd: NodeListener[];
+	signals: Map<NodeJS.Signals, NodeListener[]>;
+};
+
+function takeListenerSnapshot(): ListenerSnapshot {
+	const signals: NodeJS.Signals[] = process.platform === "win32" ? ["SIGTERM"] : ["SIGTERM", "SIGHUP"];
+	return {
+		stdinEnd: process.stdin.listeners("end") as NodeListener[],
+		signals: new Map(signals.map((signal) => [signal, process.listeners(signal) as NodeListener[]])),
+	};
+}
+
+function restoreListeners(snapshot: ListenerSnapshot): void {
+	for (const listener of process.stdin.listeners("end") as NodeListener[]) {
+		if (!snapshot.stdinEnd.includes(listener)) {
+			process.stdin.off("end", listener);
+		}
+	}
+
+	for (const [signal, previousListeners] of snapshot.signals) {
+		for (const listener of process.listeners(signal) as NodeListener[]) {
+			if (!previousListeners.includes(listener)) {
+				process.off(signal, listener);
+			}
+		}
+	}
+}
+
+function parseOutputLines(): Array<Record<string, unknown>> {
+	return rpcIo.outputLines
+		.flatMap((line) => line.split("\n"))
+		.filter((line) => line.trim().length > 0)
+		.map((line) => JSON.parse(line) as Record<string, unknown>);
+}
+
+function createRuntimeHost(harness: Harness): AgentSessionRuntime {
+	return {
+		session: harness.session,
+		newSession: vi.fn(async () => ({ cancelled: true })),
+		switchSession: vi.fn(async () => ({ cancelled: true })),
+		fork: vi.fn(async () => ({ cancelled: true, selectedText: "" })),
+		dispose: vi.fn(async () => {}),
+		setRebindSession: vi.fn(),
+	} as unknown as AgentSessionRuntime;
+}
+
+describe("RPC unknown command responses (#5868)", () => {
+	afterEach(() => {
+		rpcIo.outputLines = [];
+		rpcIo.lineHandler = undefined;
+	});
+
+	test("preserves the request id on unknown command errors", async () => {
+		const listenerSnapshot = takeListenerSnapshot();
+		const harness = await createHarness();
+
+		try {
+			void runRpcMode(createRuntimeHost(harness));
+			await vi.waitFor(() => expect(rpcIo.lineHandler).toBeDefined());
+
+			rpcIo.lineHandler?.(JSON.stringify({ id: "test", type: "foobar" }));
+
+			await vi.waitFor(() => {
+				expect(parseOutputLines()).toContainEqual({
+					id: "test",
+					type: "response",
+					command: "foobar",
+					success: false,
+					error: "Unknown command: foobar",
+				});
+			});
+		} finally {
+			harness.cleanup();
+			restoreListeners(listenerSnapshot);
+		}
+	});
+});

packages/coding-agent/test/suite/regressions/extension-factory-cache.test.ts

@@ -0,0 +1,131 @@
+import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { clearExtensionCache, loadExtensions, loadExtensionsCached } from "../../../src/core/extensions/loader.ts";
+import { DefaultResourceLoader } from "../../../src/core/resource-loader.ts";
+
+interface TestState {
+	moduleLoads?: number;
+	factoryRuns?: number;
+}
+
+function state(): TestState {
+	const global = globalThis as typeof globalThis & { __extensionFactoryCacheTest?: TestState };
+	if (!global.__extensionFactoryCacheTest) {
+		global.__extensionFactoryCacheTest = {};
+	}
+	return global.__extensionFactoryCacheTest;
+}
+
+function resetState(): void {
+	delete (globalThis as typeof globalThis & { __extensionFactoryCacheTest?: TestState }).__extensionFactoryCacheTest;
+}
+
+function writeCountingExtension(filePath: string): void {
+	writeFileSync(
+		filePath,
+		`
+const state = (globalThis.__extensionFactoryCacheTest ??= {});
+state.moduleLoads = (state.moduleLoads ?? 0) + 1;
+
+export default function () {
+	state.factoryRuns = (state.factoryRuns ?? 0) + 1;
+}
+`,
+		"utf-8",
+	);
+}
+
+describe("extension factory cache", () => {
+	const roots: string[] = [];
+
+	function fixture(name: string) {
+		const root = join(tmpdir(), `pi-extension-cache-${name}-${Date.now()}-${Math.random().toString(36).slice(2)}`);
+		const cwd = join(root, "project");
+		const agentDir = join(root, "agent");
+		mkdirSync(cwd, { recursive: true });
+		mkdirSync(agentDir, { recursive: true });
+		roots.push(root);
+		return { root, cwd, agentDir };
+	}
+
+	beforeEach(() => {
+		resetState();
+		clearExtensionCache();
+	});
+
+	afterEach(() => {
+		while (roots.length > 0) {
+			const root = roots.pop();
+			if (root && existsSync(root)) {
+				rmSync(root, { recursive: true, force: true });
+			}
+		}
+		resetState();
+		clearExtensionCache();
+	});
+
+	it("caches extension modules for cached same-cwd loads but reruns factories", async () => {
+		const { root, cwd } = fixture("same-cwd");
+		const extensionPath = join(root, "counting.ts");
+		writeCountingExtension(extensionPath);
+
+		const first = await loadExtensionsCached([extensionPath], cwd);
+		const second = await loadExtensionsCached([extensionPath], cwd);
+
+		expect(state().moduleLoads).toBe(1);
+		expect(state().factoryRuns).toBe(2);
+		expect(first.extensions[0]).not.toBe(second.extensions[0]);
+		expect(first.runtime).not.toBe(second.runtime);
+	});
+
+	it("does not cache direct loadExtensions calls", async () => {
+		const { root, cwd } = fixture("direct");
+		const extensionPath = join(root, "counting.ts");
+		writeCountingExtension(extensionPath);
+
+		await loadExtensions([extensionPath], cwd);
+		await loadExtensions([extensionPath], cwd);
+
+		expect(state().moduleLoads).toBe(2);
+		expect(state().factoryRuns).toBe(2);
+	});
+
+	it("clears the cache on resource loader reload", async () => {
+		const { cwd, agentDir } = fixture("reload");
+		const extensionDir = join(agentDir, "extensions");
+		mkdirSync(extensionDir, { recursive: true });
+		writeCountingExtension(join(extensionDir, "counting.ts"));
+		const loader = new DefaultResourceLoader({
+			cwd,
+			agentDir,
+			noSkills: true,
+			noPromptTemplates: true,
+			noThemes: true,
+		});
+
+		await loader.reload();
+		await loader.reload();
+
+		expect(state().moduleLoads).toBe(2);
+		expect(state().factoryRuns).toBe(2);
+	});
+
+	it("keeps the cache scoped to one cwd", async () => {
+		const { root } = fixture("cross-cwd");
+		const firstCwd = join(root, "first");
+		const secondCwd = join(root, "second");
+		mkdirSync(firstCwd, { recursive: true });
+		mkdirSync(secondCwd, { recursive: true });
+		const extensionPath = join(root, "counting.ts");
+		writeCountingExtension(extensionPath);
+
+		await loadExtensionsCached([extensionPath], firstCwd);
+		await loadExtensionsCached([extensionPath], secondCwd);
+		await loadExtensionsCached([extensionPath], secondCwd);
+
+		expect(state().moduleLoads).toBe(2);
+		expect(state().factoryRuns).toBe(3);
+	});
+});

packages/coding-agent/test/syntax-highlight.test.ts

@@ -1,4 +1,6 @@
-import { describe, expect, it } from "vitest";
+import { resetCapabilitiesCache, setCapabilities } from "@earendil-works/pi-tui";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { highlightCode, initTheme } from "../src/modes/interactive/theme/theme.ts";
 import { highlight, renderHighlightedHtml, supportsLanguage } from "../src/utils/syntax-highlight.ts";
 
 describe("syntax highlight renderer", () => {
@@ -46,3 +48,29 @@ describe("syntax highlight renderer", () => {
 		expect(rendered).toContain("[number:1]");
 	});
 });
+
+describe("theme syntax highlighting", () => {
+	beforeEach(() => {
+		setCapabilities({ images: null, trueColor: true, hyperlinks: false });
+		initTheme("dark");
+	});
+
+	afterEach(() => {
+		resetCapabilitiesCache();
+	});
+
+	it("colors diff additions and deletions in fenced diff blocks", () => {
+		const lines = highlightCode("-old\n+new\n", "diff");
+
+		expect(lines[0]).toBe("\x1b[38;2;204;102;102m-old\x1b[39m");
+		expect(lines[1]).toBe("\x1b[38;2;181;189;104m+new\x1b[39m");
+	});
+
+	it("keeps cli-highlight default styled scopes mapped to theme styles", () => {
+		expect(highlightCode("const re = /foo+/gi;", "javascript")[0]).toContain(
+			"\x1b[38;2;206;145;120m/foo+/gi\x1b[39m",
+		);
+		expect(highlightCode("@decorator", "python")[0]).toBe("\x1b[38;2;128;128;128m@decorator\x1b[39m");
+		expect(highlightCode("<div></div>", "html")[0]).toContain("\x1b[38;2;86;156;214mdiv\x1b[39m");
+	});
+});

packages/coding-agent/test/theme-detection.test.ts

@@ -1,24 +1,26 @@
-import { resetCapabilitiesCache, setCapabilities } from "@earendil-works/pi-tui";
+import { type RgbColor, resetCapabilitiesCache, setCapabilities } from "@earendil-works/pi-tui";
 import { afterEach, describe, expect, it } from "vitest";
 import {
-	detectTerminalBackground,
+	detectTerminalBackgroundFromEnv,
+	detectTerminalBackgroundTheme,
 	getThemeByName,
 	getThemeForRgbColor,
-	parseOsc11BackgroundColor,
+	parseAutoThemeSetting,
+	resolveThemeSetting,
 } from "../src/modes/interactive/theme/theme.ts";
 
 afterEach(() => {
 	resetCapabilitiesCache();
 });
 
-describe("detectTerminalBackground", () => {
+describe("detectTerminalBackgroundFromEnv", () => {
 	it("uses the COLORFGBG background color index", () => {
-		expect(detectTerminalBackground({ env: { COLORFGBG: "0;15" } })).toMatchObject({
+		expect(detectTerminalBackgroundFromEnv({ env: { COLORFGBG: "0;15" } })).toMatchObject({
 			theme: "light",
 			source: "COLORFGBG",
 			confidence: "high",
 		});
-		expect(detectTerminalBackground({ env: { COLORFGBG: "15;0" } })).toMatchObject({
+		expect(detectTerminalBackgroundFromEnv({ env: { COLORFGBG: "15;0" } })).toMatchObject({
 			theme: "dark",
 			source: "COLORFGBG",
 			confidence: "high",
@@ -26,11 +28,11 @@ describe("detectTerminalBackground", () => {
 	});
 
 	it("uses the last COLORFGBG field as the background", () => {
-		expect(detectTerminalBackground({ env: { COLORFGBG: "0;7;15" } }).theme).toBe("light");
+		expect(detectTerminalBackgroundFromEnv({ env: { COLORFGBG: "0;7;15" } }).theme).toBe("light");
 	});
 
 	it("defaults to dark without terminal background hints", () => {
-		expect(detectTerminalBackground({ env: {} })).toMatchObject({
+		expect(detectTerminalBackgroundFromEnv({ env: {} })).toMatchObject({
 			theme: "dark",
 			source: "fallback",
 			confidence: "low",
@@ -38,6 +40,65 @@ describe("detectTerminalBackground", () => {
 	});
 });
 
+describe("detectTerminalBackgroundTheme", () => {
+	it("uses the queried terminal background before environment hints", async () => {
+		let queriedTimeoutMs: number | undefined;
+		const detection = await detectTerminalBackgroundTheme({
+			env: { COLORFGBG: "15;0" },
+			timeoutMs: 250,
+			ui: {
+				async queryTerminalBackgroundColor({ timeoutMs }: { timeoutMs: number }): Promise<RgbColor | undefined> {
+					queriedTimeoutMs = timeoutMs;
+					return { r: 250, g: 250, b: 250 };
+				},
+			},
+		});
+
+		expect(queriedTimeoutMs).toBe(250);
+		expect(detection).toMatchObject({
+			theme: "light",
+			source: "terminal background",
+			confidence: "high",
+		});
+	});
+
+	it("falls back to environment hints when the terminal query returns no color", async () => {
+		const detection = await detectTerminalBackgroundTheme({
+			env: { COLORFGBG: "15;0" },
+			timeoutMs: 250,
+			ui: {
+				async queryTerminalBackgroundColor(): Promise<RgbColor | undefined> {
+					return undefined;
+				},
+			},
+		});
+
+		expect(detection).toMatchObject({
+			theme: "dark",
+			source: "COLORFGBG",
+			confidence: "high",
+		});
+	});
+
+	it("falls back to environment hints when the terminal query fails", async () => {
+		const detection = await detectTerminalBackgroundTheme({
+			env: { COLORFGBG: "0;15" },
+			timeoutMs: 250,
+			ui: {
+				async queryTerminalBackgroundColor(): Promise<RgbColor | undefined> {
+					throw new Error("terminal write failed");
+				},
+			},
+		});
+
+		expect(detection).toMatchObject({
+			theme: "light",
+			source: "COLORFGBG",
+			confidence: "high",
+		});
+	});
+});
+
 describe("theme color mode", () => {
 	it("uses terminal capabilities", () => {
 		setCapabilities({ images: null, trueColor: false, hyperlinks: false });
@@ -54,18 +115,19 @@ describe("theme color mode", () => {
 	});
 });
 
-describe("parseOsc11BackgroundColor", () => {
-	it("parses 16-bit OSC 11 rgb responses", () => {
-		expect(parseOsc11BackgroundColor("\x1b]11;rgb:0000/8000/ffff\x07")).toEqual({ r: 0, g: 128, b: 255 });
-	});
-
-	it("parses OSC 11 hex responses", () => {
-		expect(parseOsc11BackgroundColor("\x1b]11;#ffffff\x1b\\")).toEqual({ r: 255, g: 255, b: 255 });
-		expect(parseOsc11BackgroundColor("\x1b]11;#000000\x07")).toEqual({ r: 0, g: 0, b: 0 });
-	});
-
+describe("theme detection from RGB", () => {
 	it("classifies RGB colors by luminance", () => {
 		expect(getThemeForRgbColor({ r: 8, g: 8, b: 8 })).toBe("dark");
 		expect(getThemeForRgbColor({ r: 250, g: 250, b: 250 })).toBe("light");
 	});
 });
+
+describe("theme setting helpers", () => {
+	it("parses and resolves automatic theme settings", () => {
+		expect(parseAutoThemeSetting("light/dark")).toEqual({ lightTheme: "light", darkTheme: "dark" });
+		expect(resolveThemeSetting("dark", "light")).toBe("dark");
+		expect(resolveThemeSetting("light/dark", "light")).toBe("light");
+		expect(resolveThemeSetting("light/dark", "dark")).toBe("dark");
+		expect(resolveThemeSetting("light/dark/extra", "dark")).toBeUndefined();
+	});
+});

packages/coding-agent/test/tool-execution-component.test.ts

@@ -67,6 +67,37 @@ describe("ToolExecutionComponent parity", () => {
 		expect(rendered).toContain("custom result");
 	});
 
+	test("self-rendered empty tool rows take no layout space", () => {
+		const toolDefinition: ToolDefinition = {
+			...createBaseToolDefinition(),
+			renderShell: "self",
+			renderCall: () => new Text("", 0, 0),
+			renderResult: () => new Text("", 0, 0),
+		};
+
+		const component = new ToolExecutionComponent(
+			"custom_tool",
+			"tool-empty-self-render",
+			{},
+			{},
+			toolDefinition,
+			createFakeTui(),
+			process.cwd(),
+		);
+		expect(component.render(120)).toEqual([]);
+
+		component.updateResult(
+			{
+				content: [],
+				details: {},
+				isError: false,
+			},
+			false,
+		);
+
+		expect(component.render(120)).toEqual([]);
+	});
+
 	test("uses built-in rendering for built-in overrides without custom renderers", () => {
 		const overrideDefinition: ToolDefinition = {
 			...createBaseToolDefinition("edit"),
@@ -191,6 +222,7 @@ describe("ToolExecutionComponent parity", () => {
 			process.cwd(),
 		);
 		component.updateResult({ content: [{ type: "text", text: "hello" }], details: undefined, isError: false }, false);
+		component.setExpanded(true);
 		const rendered = stripAnsi(component.render(120).join("\n"));
 		expect(rendered).toContain("override call");
 		expect(rendered).toContain("hello");
@@ -362,12 +394,38 @@ describe("ToolExecutionComponent parity", () => {
 			{ content: [{ type: "text", text: "one\ntwo\n" }], details: undefined, isError: false },
 			false,
 		);
+		component.setExpanded(true);
 		const rendered = stripAnsi(component.render(120).join("\n"));
 		expect(rendered).toContain("one");
 		expect(rendered).toContain("two");
 		expect(rendered).not.toContain("two\n\n");
 	});
 
+	test("collapses ordinary read results until expanded", () => {
+		const component = new ToolExecutionComponent(
+			"read",
+			"tool-ordinary-read-collapsed",
+			{ path: "notes.txt" },
+			{},
+			createReadToolDefinition(process.cwd()),
+			createFakeTui(),
+			process.cwd(),
+		);
+		component.updateResult(
+			{ content: [{ type: "text", text: "hidden content" }], details: undefined, isError: false },
+			false,
+		);
+
+		const collapsed = stripAnsi(component.render(120).join("\n"));
+		expect(collapsed).toContain("read");
+		expect(collapsed).toContain("notes.txt");
+		expect(collapsed).not.toContain("hidden content");
+
+		component.setExpanded(true);
+		const expanded = stripAnsi(component.render(120).join("\n"));
+		expect(expanded).toContain("hidden content");
+	});
+
 	for (const scenario of [
 		{
 			title: "SKILL.md",

packages/coding-agent/test/tools.test.ts

@@ -44,6 +44,7 @@ describe("Coding Agent Tools", () => {
 	});
 
 	afterEach(() => {
+		vi.restoreAllMocks();
 		// Clean up test directory
 		rmSync(testDir, { recursive: true, force: true });
 	});
@@ -535,6 +536,56 @@ describe("Coding Agent Tools", () => {
 			expect(getShellConfigSpy).toHaveBeenCalledWith("/custom/bash");
 		});
 
+		it("should send commands over stdin when shell resolution requires it", async () => {
+			vi.spyOn(shellModule, "getShellConfig").mockReturnValue({
+				shell: process.execPath,
+				args: [
+					"-e",
+					'let input = ""; process.stdin.setEncoding("utf8"); process.stdin.on("data", (chunk) => { input += chunk; }); process.stdin.on("end", () => { process.stdout.write(input); });',
+				],
+				commandTransport: "stdin",
+			});
+			const chunks: Buffer[] = [];
+			const ops = createLocalBashOperations({ shellPath: "C:\\Windows\\System32\\bash.exe" });
+			const nameExpansion = "$" + "{name}";
+			const countExpansion = "$" + "{count}";
+			const iExpansion = "$" + "{i}";
+			const command = `name='World'; echo "Hello, ${nameExpansion}!"; count=3; for i in $(seq 1 ${countExpansion}); do echo "Iteration ${iExpansion} of ${countExpansion}"; done`;
+
+			const result = await ops.exec(command, testDir, {
+				onData: (data) => chunks.push(data),
+			});
+
+			expect(result.exitCode).toBe(0);
+			expect(Buffer.concat(chunks).toString("utf-8")).toBe(command);
+		});
+
+		it("should resolve legacy WSL bash.exe to stdin command transport", () => {
+			if (process.platform === "win32") return;
+			const originalCwd = process.cwd();
+			const platformDescriptor = Object.getOwnPropertyDescriptor(process, "platform");
+			const shellPath = "C:\\Windows\\System32\\bash.exe";
+			writeFileSync(join(testDir, shellPath), "");
+			try {
+				process.chdir(testDir);
+				Object.defineProperty(process, "platform", {
+					configurable: true,
+					value: "win32",
+				});
+
+				expect(shellModule.getShellConfig(shellPath)).toEqual({
+					shell: shellPath,
+					args: ["-s"],
+					commandTransport: "stdin",
+				});
+			} finally {
+				process.chdir(originalCwd);
+				if (platformDescriptor) {
+					Object.defineProperty(process, "platform", platformDescriptor);
+				}
+			}
+		});
+
 		it("should prepend command prefix when configured", async () => {
 			const bashWithPrefix = createBashTool(testDir, {
 				commandPrefix: "export TEST_VAR=hello",
@@ -980,6 +1031,57 @@ describe("edit tool fuzzy matching", () => {
 
 		expect(readFileSync(testFile, "utf-8")).toBe("console.log('world');\nhello universe\n");
 	});
+
+	it("should preserve the correct occurrence when fuzzy replacement equals a nearby line", async () => {
+		const testFile = join(testDir, "fuzzy-preserve-duplicate-line.txt");
+		const originalContent = ["replace me\u0020\u0020\u0020", "after\u0020\u0020\u0020", ""].join("\n");
+		writeFileSync(testFile, originalContent);
+
+		const result = await editTool.execute("test-fuzzy-preserve-duplicate-line", {
+			path: testFile,
+			edits: [{ oldText: "replace me\n", newText: "after\n" }],
+		});
+
+		const expectedContent = ["after", "after\u0020\u0020\u0020", ""].join("\n");
+		expect(readFileSync(testFile, "utf-8")).toBe(expectedContent);
+		expect(applyPatch(originalContent, result.details?.patch ?? "")).toBe(expectedContent);
+	});
+
+	it("should preserve untouched lines and produce an applicable patch for fuzzy multi-edits", async () => {
+		const testFile = join(testDir, "fuzzy-preserve-multi.txt");
+		const originalContent = [
+			"keep before\u0020\u0020",
+			"first target\u0020\u0020",
+			"first after",
+			"keep middle\u0020\u0020\u0020",
+			"second target\u0020\u0020",
+			"second after",
+			"keep after\u0020\u0020",
+			"",
+		].join("\n");
+		writeFileSync(testFile, originalContent);
+
+		const result = await editTool.execute("test-fuzzy-preserve-multi", {
+			path: testFile,
+			edits: [
+				{ oldText: "first target\nfirst after", newText: "FIRST\nFIRST2" },
+				{ oldText: "second target\nsecond after", newText: "SECOND\nSECOND2" },
+			],
+		});
+
+		const expectedContent = [
+			"keep before\u0020\u0020",
+			"FIRST",
+			"FIRST2",
+			"keep middle\u0020\u0020\u0020",
+			"SECOND",
+			"SECOND2",
+			"keep after\u0020\u0020",
+			"",
+		].join("\n");
+		expect(readFileSync(testFile, "utf-8")).toBe(expectedContent);
+		expect(applyPatch(originalContent, result.details?.patch ?? "")).toBe(expectedContent);
+	});
 });
 
 describe("edit tool CRLF handling", () => {

packages/coding-agent/test/tree-selector.test.ts

@@ -1,4 +1,5 @@
-import { setKeybindings } from "@earendil-works/pi-tui";
+import { stripVTControlCharacters } from "node:util";
+import { setKeybindings, visibleWidth } from "@earendil-works/pi-tui";
 import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { KeybindingsManager } from "../src/core/keybindings.ts";
 import type {
@@ -248,6 +249,29 @@ describe("TreeSelectorComponent", () => {
 		});
 	});
 
+	describe("help", () => {
+		test("renders semantic help rows without truncating narrow terminal controls", () => {
+			const entries = [userMessage("user-1", null, "hello"), assistantMessage("asst-1", "user-1", "hi")];
+			const tree = buildTree(entries);
+			const selector = new TreeSelectorComponent(
+				tree,
+				"asst-1",
+				24,
+				() => {},
+				() => {},
+			);
+
+			const plainLines = selector.render(30).map(stripVTControlCharacters);
+			const plain = plainLines.join("\n");
+			expect(plain).toContain("branch");
+			expect(plain).toContain("filters");
+			expect(plain).toContain("cycle");
+			expect(plain).toContain("label time");
+			expect(plain).not.toContain("...");
+			expect(plainLines.every((line) => visibleWidth(line) <= 30)).toBe(true);
+		});
+	});
+
 	describe("label timestamps", () => {
 		test("toggles label timestamps for labeled nodes", () => {
 			const entries = [userMessage("user-1", null, "hello"), assistantMessage("asst-1", "user-1", "hi")];

packages/coding-agent/test/trigger-compact-extension.test.ts

@@ -4,6 +4,7 @@ import type { ExtensionAPI, ExtensionCommandContext, ExtensionContext } from "..
 
 function createContext(tokens: number | null, compact = vi.fn()): ExtensionContext {
 	return {
+		mode: "print",
 		hasUI: false,
 		ui: {} as ExtensionContext["ui"],
 		cwd: process.cwd(),
@@ -11,6 +12,7 @@ function createContext(tokens: number | null, compact = vi.fn()): ExtensionConte
 		modelRegistry: {} as ExtensionContext["modelRegistry"],
 		model: undefined,
 		isIdle: () => true,
+		isProjectTrusted: () => true,
 		signal: undefined,
 		abort: vi.fn(),
 		hasPendingMessages: () => false,

packages/coding-agent/test/trust-manager.test.ts

@@ -0,0 +1,67 @@
+import { mkdirSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { hasTrustRequiringProjectResources, ProjectTrustStore } from "../src/core/trust-manager.ts";
+
+describe("ProjectTrustStore", () => {
+	let tempDir: string;
+	let agentDir: string;
+	let cwd: string;
+
+	beforeEach(() => {
+		tempDir = join(tmpdir(), `trust-test-${Date.now()}-${Math.random().toString(36).slice(2)}`);
+		agentDir = join(tempDir, "agent");
+		cwd = join(tempDir, "project");
+		mkdirSync(agentDir, { recursive: true });
+		mkdirSync(cwd, { recursive: true });
+	});
+
+	afterEach(() => {
+		rmSync(tempDir, { recursive: true, force: true });
+	});
+
+	it("stores decisions and inherits from parent directories", () => {
+		const store = new ProjectTrustStore(agentDir);
+		const parentDir = join(tempDir, "trusted-parent");
+		const childDir = join(parentDir, "project");
+		mkdirSync(childDir, { recursive: true });
+
+		expect(store.get(childDir)).toBeNull();
+		store.set(parentDir, true);
+		expect(store.get(childDir)).toBe(true);
+		store.set(childDir, false);
+		expect(store.get(childDir)).toBe(false);
+		store.set(childDir, null);
+		expect(store.get(childDir)).toBe(true);
+	});
+
+	it("detects trust-requiring project resources", () => {
+		const originalHome = process.env.HOME;
+		process.env.HOME = tempDir;
+		try {
+			mkdirSync(join(tempDir, ".pi", "agent"), { recursive: true });
+			mkdirSync(join(tempDir, ".agents", "skills"), { recursive: true });
+			expect(hasTrustRequiringProjectResources(tempDir)).toBe(false);
+			expect(hasTrustRequiringProjectResources(cwd)).toBe(false);
+
+			writeFileSync(join(tempDir, ".pi", "settings.json"), "{}");
+			expect(hasTrustRequiringProjectResources(tempDir)).toBe(true);
+			rmSync(join(tempDir, ".pi", "settings.json"), { force: true });
+
+			mkdirSync(join(cwd, ".pi"), { recursive: true });
+			writeFileSync(join(cwd, ".pi", "settings.json"), "{}");
+			expect(hasTrustRequiringProjectResources(cwd)).toBe(true);
+
+			rmSync(join(cwd, ".pi"), { recursive: true, force: true });
+			mkdirSync(join(cwd, ".agents", "skills"), { recursive: true });
+			expect(hasTrustRequiringProjectResources(cwd)).toBe(true);
+		} finally {
+			if (originalHome === undefined) {
+				delete process.env.HOME;
+			} else {
+				process.env.HOME = originalHome;
+			}
+		}
+	});
+});

packages/coding-agent/test/trust-selector.test.ts

@@ -0,0 +1,87 @@
+import { setKeybindings } from "@earendil-works/pi-tui";
+import { beforeAll, beforeEach, describe, expect, it, vi } from "vitest";
+import { KeybindingsManager } from "../src/core/keybindings.ts";
+import { TrustSelectorComponent } from "../src/modes/interactive/components/trust-selector.ts";
+import { initTheme } from "../src/modes/interactive/theme/theme.ts";
+import { stripAnsi } from "../src/utils/ansi.ts";
+
+describe("TrustSelectorComponent", () => {
+	beforeAll(() => {
+		initTheme("dark");
+	});
+
+	beforeEach(() => {
+		setKeybindings(new KeybindingsManager());
+	});
+
+	it("marks the saved trusted decision", () => {
+		const selector = new TrustSelectorComponent({
+			cwd: "/project",
+			savedDecision: { path: "/project", decision: true },
+			projectTrusted: true,
+			onSelect: () => {},
+			onCancel: () => {},
+		});
+
+		const output = stripAnsi(selector.render(120).join("\n"));
+
+		expect(output).toContain("Saved decision: trusted (/project)");
+		expect(output).toContain("Current session: trusted");
+		expect(output).toContain("Trust ✓");
+		expect(output).not.toContain("Do not trust ✓");
+	});
+
+	it("selects a trust decision", () => {
+		const onSelect = vi.fn();
+		const selector = new TrustSelectorComponent({
+			cwd: "/project",
+			savedDecision: null,
+			projectTrusted: false,
+			onSelect,
+			onCancel: () => {},
+		});
+
+		selector.handleInput("\n");
+
+		expect(onSelect).toHaveBeenCalledWith({ trusted: true, updates: [{ path: "/project", decision: true }] });
+	});
+
+	it("labels saved ancestor decisions as inherited", () => {
+		const selector = new TrustSelectorComponent({
+			cwd: "/parent/project/nested",
+			savedDecision: { path: "/parent", decision: true },
+			projectTrusted: true,
+			onSelect: () => {},
+			onCancel: () => {},
+		});
+
+		const output = stripAnsi(selector.render(120).join("\n"));
+
+		expect(output).toContain("Saved decision: trusted (inherited from /parent)");
+	});
+
+	it("adds a trust parent option", () => {
+		const onSelect = vi.fn();
+		const selector = new TrustSelectorComponent({
+			cwd: "/parent/project",
+			savedDecision: { path: "/parent", decision: true },
+			projectTrusted: true,
+			onSelect,
+			onCancel: () => {},
+		});
+
+		const output = stripAnsi(selector.render(120).join("\n"));
+		expect(output).toContain("Saved decision: trusted (inherited from /parent)");
+		expect(output).toContain("Trust parent folder (/parent) ✓");
+
+		selector.handleInput("\n");
+
+		expect(onSelect).toHaveBeenCalledWith({
+			trusted: true,
+			updates: [
+				{ path: "/parent", decision: true },
+				{ path: "/parent/project", decision: null },
+			],
+		});
+	});
+});

packages/coding-agent/test/version-check.test.ts

@@ -29,6 +29,7 @@ describe("version checks", () => {
 		expect(comparePackageVersions("0.70.6", "0.70.5")).toBeGreaterThan(0);
 		expect(comparePackageVersions("0.70.5", "0.70.5")).toBe(0);
 		expect(comparePackageVersions("0.70.4", "0.70.5")).toBeLessThan(0);
+		expect(comparePackageVersions("5.0.0-beta.20", "5.0.0-beta.9")).toBeGreaterThan(0);
 		expect(isNewerPackageVersion("0.70.5", "0.70.5")).toBe(false);
 		expect(isNewerPackageVersion("0.70.6", "0.70.5")).toBe(true);
 	});