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>
2026-06-21 20:18:02 +08:00
parent 131546b2fe d93b92baca
commit 8ad811c977
409 changed files with 27402 additions and 6514 deletions
--- a/packages/agent/CHANGELOG.md
+++ b/packages/agent/CHANGELOG.md
@@ -2,6 +2,64 @@

 ## [Unreleased]

+## [0.79.9] - 2026-06-20
+
+### Fixed
+
+- Fixed Node execution environment commands through legacy WSL `bash.exe` to pass scripts over stdin so shell variables expand in the target bash ([#5893](https://github.com/earendil-works/pi/issues/5893)).
+
+## [0.79.8] - 2026-06-19
+
+### Added
+
+- Added `@earendil-works/pi-agent-core/base` for bundlers that want to pair the agent core with selective `@earendil-works/pi-ai/base` provider registration ([#5348](https://github.com/earendil-works/pi/pull/5348) by [@FredKSchott](https://github.com/FredKSchott)).
+
+## [0.79.7] - 2026-06-18
+
+## [0.79.6] - 2026-06-16
+
+## [0.79.5] - 2026-06-16
+
+## [0.79.4] - 2026-06-15
+
+## [0.79.3] - 2026-06-13
+
+## [0.79.2] - 2026-06-12
+
+### Fixed
+
+- Fixed late tool progress callbacks after tool settlement to be ignored instead of emitting stale `tool_execution_update` events ([#5573](https://github.com/earendil-works/pi/issues/5573)).
+
+## [0.79.1] - 2026-06-09
+
+## [0.79.0] - 2026-06-08
+
+### Fixed
+
+- Fixed the compaction summarization system prompt to use neutral AI assistant wording for non-coding agents ([#5401](https://github.com/earendil-works/pi/issues/5401)).
+
+## [0.78.1] - 2026-06-04
+
+## [0.78.0] - 2026-05-29
+
+## [0.77.0] - 2026-05-28
+
+### Breaking Changes
+
+- Renamed agent harness `model_select` and `thinking_level_select` events to `model_update` and `thinking_level_update`.
+
+### Added
+
+- Added agent harness tool registry APIs, `tools_update` events, branch-scoped active-tool persistence, and duplicate tool validation.
+
+## [0.76.0] - 2026-05-27
+
+### Fixed
+
+- Fixed context token estimates to count user image attachments consistently with tool result images ([#4983](https://github.com/earendil-works/pi/issues/4983)).
+
+## [0.75.5] - 2026-05-23
+
 ## [0.75.4] - 2026-05-20

 ### Changed
--- a/packages/agent/README.md
+++ b/packages/agent/README.md
@@ -31,6 +31,24 @@ agent.subscribe((event) => {
 await agent.prompt("Hello!");
 ```

+## Base Entry Point
+
+Use `@earendil-works/pi-agent-core/base` with `@earendil-works/pi-ai/base` when bundling applications that should include only selected provider transports:
+
+```typescript
+import { Agent } from "@earendil-works/pi-agent-core/base";
+import { getModel } from "@earendil-works/pi-ai/base";
+import { register } from "@earendil-works/pi-ai/anthropic";
+
+register();
+
+const agent = new Agent({
+  initialState: { model: getModel("anthropic", "claude-sonnet-4-6") },
+});
+```
+
+The default `@earendil-works/pi-agent-core` entry point remains batteries-included and registers pi-ai's lazy built-in transports for backward compatibility.
+
 ## Core Concepts

 ### AgentMessage vs LLM Message
--- a/packages/agent/docs/agent-harness.md
+++ b/packages/agent/docs/agent-harness.md
@@ -258,13 +258,14 @@ Done:
 - Exported `QueueMode` from core types.
 - Added `AgentHarnessOptions.steeringMode` and `followUpMode`.
 - Added live `getSteeringMode()` / `setSteeringMode()` and `getFollowUpMode()` / `setFollowUpMode()`.
+- Added `getTools()` and `getActiveTools()`.
+- Added `tools_update` observability events, including active-tool-only updates.
+- Active tool changes are persisted as branch-scoped `active_tools_change` entries.
+- Duplicate tool names and duplicate active tool names reject.

 Remaining:

- Add `getTools()` semantics.
- Add `getActiveTools()` semantics.
- Decide and implement tool update observability events.
- Include active-tool-only updates in the runtime config observability plan.
+- None.

 Notes:

--- a/packages/agent/docs/durable-harness.md
+++ b/packages/agent/docs/durable-harness.md
@@ -14,6 +14,8 @@ A fully durable `AgentHarness` is not realistic by itself because important depe
 - resource loaders
 - system-prompt callbacks/modifiers

+Tool registries are runtime dependencies. The harness should persist serializable tool configuration, such as active tool names, but not concrete tool implementations.
+
 The practical target is a semi-durable harness:

 - session is the durable append-only state tree
@@ -29,6 +31,7 @@ Existing session state already includes harness state:

 - model changes
 - thinking-level changes
+- active-tool changes
 - leaf entries
 - labels
 - compactions and branch summaries
@@ -50,10 +53,38 @@ The app must recreate compatible runtime dependencies:

 Harness can validate stable IDs/versions/hashes when available, but it cannot serialize these dependencies itself.

+## Runtime configuration and restore
+
+Constructor options remain explicit runtime configuration and do not read session state. Hidden async restore in a constructor would make failure handling ambiguous.
+
+A future async builder/factory should own durable restore:
+
+```ts
+const harness = await AgentHarness.builder()
+  .env(env)
+  .session(session)
+  .model(defaultModel)
+  .tools(runtimeTools)
+  .defaultActiveTools(["read", "edit"])
+  .restore({ missingActiveTools: "fail" });
+```
+
+`restore()` should read the active branch, reduce durable harness configuration, apply defaults for missing entries, validate against app-supplied runtime dependencies, construct the harness, and optionally emit `source: "restore"` update events after construction.
+
+For active tools:
+
+- `active_tools_change` entries are branch-scoped durable config.
+- If no `active_tools_change` exists on the branch, restore uses builder defaults, or all registered tools if no default active names were supplied.
+- Active tool names must be unique.
+- Tool registry names must be unique.
+- Missing restored active tool names should fail restore by default; permissive drop/disable policies can be added explicitly later.
+- Concrete tools are never restored from session; the host app must provide compatible tools.
+
 ## What harness should persist

 Minimum useful durability entries:

+- branch-scoped active tool names
 - queued steer/followUp/nextTurn messages
 - queue consumption tied to a turn
 - pending session writes accepted during active operations
@@ -93,11 +124,11 @@ On startup:
 3. Harness reduces session entries into:
   - current leaf
   - conversation branch
-   - harness config
+   - harness config, including active tool names
   - queues
   - pending writes
   - active operation/turn/tool state
-4. Harness validates required runtime dependencies.
+4. Harness validates required runtime dependencies, including restored active tool names against the app-provided tool registry.
 5. Harness reconciles unfinished operation state.

 Provider streams are not resumable. Recovery can only retry from a durable boundary or mark the operation interrupted.
@@ -173,7 +204,7 @@ recovery: "mark_interrupted" | "retry_unfinished"

 ## Open questions

- Which harness config entries should move into session first: tools, active tools, resources, stream options, system prompt refs?
+- Which remaining harness config entries should move into session first: resources, stream options, system prompt refs?
 - Should resolved system prompt text be snapshotted per turn for audit/debug?
 - Do we require strict dependency ID/version matching on resume?
 - How much provider request data should be journaled?
--- a/packages/agent/package.json
+++ b/packages/agent/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@earendil-works/pi-agent-core",
-	"version": "0.75.4",
+	"version": "0.79.9",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -10,6 +10,10 @@
 			"types": "./dist/index.d.ts",
 			"import": "./dist/index.js"
 		},
+		"./base": {
+			"types": "./dist/base.d.ts",
+			"import": "./dist/base.js"
+		},
 		"./node": {
 			"types": "./dist/node.d.ts",
 			"import": "./dist/node.js"
@@ -29,7 +33,7 @@
 		"prepublishOnly": "npm run clean && npm run build"
 	},
 	"dependencies": {
-		"@earendil-works/pi-ai": "^0.75.4",
+		"@earendil-works/pi-ai": "^0.79.9",
 		"ignore": "7.0.5",
 		"typebox": "1.1.38",
 		"yaml": "2.9.0"
@@ -45,7 +49,7 @@
 	"license": "MIT",
 	"repository": {
 		"type": "git",
-		"url": "git+https://github.com/earendil-works/pi-mono.git",
+		"url": "git+https://github.com/earendil-works/pi.git",
 		"directory": "packages/agent"
 	},
 	"engines": {
@@ -53,8 +57,8 @@
 	},
 	"devDependencies": {
 		"@types/node": "24.12.4",
-		"@vitest/coverage-v8": "3.2.4",
+		"@vitest/coverage-v8": "4.1.9",
 		"typescript": "5.9.3",
-		"vitest": "3.2.4"
+		"vitest": "4.1.9"
 	}
 }
--- a/packages/agent/src/agent-loop.ts
+++ b/packages/agent/src/agent-loop.ts
@@ -10,7 +10,7 @@ import {
 	streamSimple,
 	type ToolResultMessage,
 	validateToolArguments,
-} from "@earendil-works/pi-ai";
+} from "@earendil-works/pi-ai/base";
 import type {
 	AgentContext,
 	AgentEvent,
@@ -631,6 +631,7 @@ async function executePreparedToolCall(
 	emit: AgentEventSink,
 ): Promise<ExecutedToolCallOutcome> {
 	const updateEvents: Promise<void>[] = [];
+	let acceptingUpdates = true;

 	try {
 		const result = await prepared.tool.execute(
@@ -638,6 +639,7 @@ async function executePreparedToolCall(
 			prepared.args as never,
 			signal,
 			(partialResult) => {
+				if (!acceptingUpdates) return;
 				updateEvents.push(
 					Promise.resolve(
 						emit({
@@ -651,14 +653,18 @@ async function executePreparedToolCall(
 				);
 			},
 		);
+		acceptingUpdates = false;
 		await Promise.all(updateEvents);
 		return { result, isError: false };
 	} catch (error) {
+		acceptingUpdates = false;
 		await Promise.all(updateEvents);
 		return {
 			result: createErrorToolResult(error instanceof Error ? error.message : String(error)),
 			isError: true,
 		};
+	} finally {
+		acceptingUpdates = false;
 	}
 }

--- a/packages/agent/src/agent.ts
+++ b/packages/agent/src/agent.ts
@@ -7,7 +7,7 @@ import {
 	type TextContent,
 	type ThinkingBudgets,
 	type Transport,
-} from "@earendil-works/pi-ai";
+} from "@earendil-works/pi-ai/base";
 import { runAgentLoop, runAgentLoopContinue } from "./agent-loop.ts";
 import type {
 	AfterToolCallContext,
--- a/packages/agent/src/base.ts
+++ b/packages/agent/src/base.ts
@@ -0,0 +1,39 @@
+export * from "./agent.ts";
+export * from "./agent-loop.ts";
+export * from "./harness/agent-harness.ts";
+export {
+	type BranchPreparation,
+	type BranchSummaryDetails,
+	type CollectEntriesResult,
+	collectEntriesForBranchSummary,
+	generateBranchSummary,
+	prepareBranchEntries,
+} from "./harness/compaction/branch-summarization.ts";
+export {
+	calculateContextTokens,
+	compact,
+	DEFAULT_COMPACTION_SETTINGS,
+	estimateContextTokens,
+	estimateTokens,
+	findCutPoint,
+	findTurnStartIndex,
+	generateSummary,
+	getLastAssistantUsage,
+	prepareCompaction,
+	serializeConversation,
+	shouldCompact,
+} from "./harness/compaction/compaction.ts";
+export * from "./harness/messages.ts";
+export * from "./harness/prompt-templates.ts";
+export * from "./harness/session/jsonl-repo.ts";
+export * from "./harness/session/memory-repo.ts";
+export * from "./harness/session/repo-utils.ts";
+export * from "./harness/session/session.ts";
+export { uuidv7 } from "./harness/session/uuid.ts";
+export * from "./harness/skills.ts";
+export * from "./harness/system-prompt.ts";
+export * from "./harness/types.ts";
+export * from "./harness/utils/shell-output.ts";
+export * from "./harness/utils/truncate.ts";
+export * from "./proxy.ts";
+export * from "./types.ts";
--- a/packages/agent/src/harness/agent-harness.ts
+++ b/packages/agent/src/harness/agent-harness.ts
@@ -4,7 +4,7 @@ import {
 	type Model,
 	streamSimple,
 	type UserMessage,
-} from "@earendil-works/pi-ai";
+} from "@earendil-works/pi-ai/base";
 import { runAgentLoop } from "../agent-loop.ts";
 import type {
 	AgentContext,
@@ -86,6 +86,16 @@ function mergeHeaders(...headers: Array<Record<string, string> | undefined>): Re
 	return hasHeaders ? merged : undefined;
 }

+function findDuplicateNames(names: string[]): string[] {
+	const seen = new Set<string>();
+	const duplicates = new Set<string>();
+	for (const name of names) {
+		if (seen.has(name)) duplicates.add(name);
+		seen.add(name);
+	}
+	return [...duplicates];
+}
+
 function applyStreamOptionsPatch(
 	base: AgentHarnessStreamOptions,
 	patch?: AgentHarnessStreamOptionsPatch,
@@ -194,12 +204,20 @@ export class AgentHarness<
 		this.streamOptions = cloneStreamOptions(options.streamOptions);
 		this.systemPrompt = options.systemPrompt;
 		this.getApiKeyAndHeaders = options.getApiKeyAndHeaders;
+		this.validateUniqueNames(
+			(options.tools ?? []).map((tool) => tool.name),
+			"Duplicate tool name(s)",
+		);
 		for (const tool of options.tools ?? []) {
 			this.tools.set(tool.name, tool);
 		}
 		this.model = options.model;
 		this.thinkingLevel = options.thinkingLevel ?? "off";
-		this.activeToolNames = options.activeToolNames ?? (options.tools ?? []).map((tool) => tool.name);
+		this.activeToolNames = options.activeToolNames
+			? [...options.activeToolNames]
+			: (options.tools ?? []).map((tool) => tool.name);
+		this.validateUniqueNames(this.activeToolNames, "Duplicate active tool name(s)");
+		this.validateToolNames(this.activeToolNames);
 		this.steeringQueueMode = options.steeringMode ?? "one-at-a-time";
 		this.followUpQueueMode = options.followUpMode ?? "one-at-a-time";
 	}
@@ -451,7 +469,14 @@ export class AgentHarness<
 		};
 	}

+	private validateUniqueNames(names: string[], message: string): void {
+		const duplicates = findDuplicateNames(names);
+		if (duplicates.length > 0)
+			throw new AgentHarnessError("invalid_argument", `${message}: ${duplicates.join(", ")}`);
+	}
+
 	private validateToolNames(toolNames: string[], tools: Map<string, TTool> = this.tools): void {
+		this.validateUniqueNames(toolNames, "Duplicate active tool name(s)");
 		const missing = toolNames.filter((name) => !tools.has(name));
 		if (missing.length > 0) throw new AgentHarnessError("invalid_argument", `Unknown tool(s): ${missing.join(", ")}`);
 	}
@@ -465,6 +490,8 @@ export class AgentHarness<
 				await this.session.appendModelChange(write.provider, write.modelId);
 			} else if (write.type === "thinking_level_change") {
 				await this.session.appendThinkingLevelChange(write.thinkingLevel);
+			} else if (write.type === "active_tools_change") {
+				await this.session.appendActiveToolsChange(write.activeToolNames);
 			} else if (write.type === "custom") {
 				await this.session.appendCustomEntry(write.customType, write.data);
 			} else if (write.type === "custom_message") {
@@ -838,10 +865,6 @@ export class AgentHarness<
 		return this.model;
 	}

-	getThinkingLevel(): ThinkingLevel {
-		return this.thinkingLevel;
-	}
-
 	async setModel(model: Model<any>): Promise<void> {
 		try {
 			const previousModel = this.model;
@@ -851,12 +874,16 @@ export class AgentHarness<
 				this.pendingSessionWrites.push({ type: "model_change", provider: model.provider, modelId: model.id });
 			}
 			this.model = model;
-			await this.emitOwn({ type: "model_select", model, previousModel, source: "set" });
+			await this.emitOwn({ type: "model_update", model, previousModel, source: "set" });
 		} catch (error) {
 			throw normalizeHarnessError(error, "session");
 		}
 	}

+	getThinkingLevel(): ThinkingLevel {
+		return this.thinkingLevel;
+	}
+
 	async setThinkingLevel(level: ThinkingLevel): Promise<void> {
 		try {
 			const previousLevel = this.thinkingLevel;
@@ -866,16 +893,70 @@ export class AgentHarness<
 				this.pendingSessionWrites.push({ type: "thinking_level_change", thinkingLevel: level });
 			}
 			this.thinkingLevel = level;
-			await this.emitOwn({ type: "thinking_level_select", level, previousLevel });
+			await this.emitOwn({ type: "thinking_level_update", level, previousLevel });
 		} catch (error) {
 			throw normalizeHarnessError(error, "session");
 		}
 	}

+	getTools(): TTool[] {
+		return [...this.tools.values()];
+	}
+
+	async setTools(tools: TTool[], activeToolNames?: string[]): Promise<void> {
+		try {
+			this.validateUniqueNames(
+				tools.map((tool) => tool.name),
+				"Duplicate tool name(s)",
+			);
+			const nextTools = new Map(tools.map((tool) => [tool.name, tool]));
+			const nextActiveToolNames = activeToolNames ? [...activeToolNames] : this.activeToolNames;
+			this.validateToolNames(nextActiveToolNames, nextTools);
+			const previousToolNames = [...this.tools.keys()];
+			const previousActiveToolNames = [...this.activeToolNames];
+			if (this.phase === "idle") {
+				await this.session.appendActiveToolsChange(nextActiveToolNames);
+			} else {
+				this.pendingSessionWrites.push({ type: "active_tools_change", activeToolNames: [...nextActiveToolNames] });
+			}
+			this.tools = nextTools;
+			this.activeToolNames = [...nextActiveToolNames];
+			await this.emitOwn({
+				type: "tools_update",
+				toolNames: [...this.tools.keys()],
+				previousToolNames,
+				activeToolNames: [...this.activeToolNames],
+				previousActiveToolNames,
+				source: "set",
+			});
+		} catch (error) {
+			throw normalizeHarnessError(error, "invalid_argument");
+		}
+	}
+
+	getActiveTools(): TTool[] {
+		return this.activeToolNames.map((name) => this.tools.get(name)!);
+	}
+
 	async setActiveTools(toolNames: string[]): Promise<void> {
 		try {
 			this.validateToolNames(toolNames);
+			const previousToolNames = [...this.tools.keys()];
+			const previousActiveToolNames = [...this.activeToolNames];
+			if (this.phase === "idle") {
+				await this.session.appendActiveToolsChange(toolNames);
+			} else {
+				this.pendingSessionWrites.push({ type: "active_tools_change", activeToolNames: [...toolNames] });
+			}
 			this.activeToolNames = [...toolNames];
+			await this.emitOwn({
+				type: "tools_update",
+				toolNames: [...this.tools.keys()],
+				previousToolNames,
+				activeToolNames: [...this.activeToolNames],
+				previousActiveToolNames,
+				source: "set",
+			});
 		} catch (error) {
 			throw normalizeHarnessError(error, "invalid_argument");
 		}
@@ -921,18 +1002,6 @@ export class AgentHarness<
 		this.streamOptions = cloneStreamOptions(streamOptions);
 	}

-	async setTools(tools: TTool[], activeToolNames?: string[]): Promise<void> {
-		try {
-			const nextTools = new Map(tools.map((tool) => [tool.name, tool]));
-			const nextActiveToolNames = activeToolNames ? [...activeToolNames] : this.activeToolNames;
-			this.validateToolNames(nextActiveToolNames, nextTools);
-			this.tools = nextTools;
-			this.activeToolNames = [...nextActiveToolNames];
-		} catch (error) {
-			throw normalizeHarnessError(error, "invalid_argument");
-		}
-	}
-
 	async abort(): Promise<AbortResult> {
 		const clearedSteer = [...this.steerQueue];
 		const clearedFollowUp = [...this.followUpQueue];
--- a/packages/agent/src/harness/compaction/branch-summarization.ts
+++ b/packages/agent/src/harness/compaction/branch-summarization.ts
@@ -1,5 +1,4 @@
-import type { Model } from "@earendil-works/pi-ai";
-import { completeSimple } from "@earendil-works/pi-ai";
+import { completeSimple, type Model } from "@earendil-works/pi-ai/base";
 import type { AgentMessage } from "../../types.ts";
 import {
 	convertToLlm,
@@ -112,6 +111,7 @@ function getMessageFromEntry(entry: SessionTreeEntry): AgentMessage | undefined
 			return createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);
 		case "thinking_level_change":
 		case "model_change":
+		case "active_tools_change":
 		case "custom":
 		case "label":
 		case "session_info":
--- a/packages/agent/src/harness/compaction/compaction.ts
+++ b/packages/agent/src/harness/compaction/compaction.ts
@@ -1,5 +1,11 @@
-import type { AssistantMessage, ImageContent, Model, TextContent, Usage } from "@earendil-works/pi-ai";
-import { completeSimple } from "@earendil-works/pi-ai";
+import {
+	type AssistantMessage,
+	completeSimple,
+	type ImageContent,
+	type Model,
+	type TextContent,
+	type Usage,
+} from "@earendil-works/pi-ai/base";
 import type { AgentMessage, ThinkingLevel } from "../../types.ts";
 import {
 	convertToLlm,
@@ -198,22 +204,33 @@ export function shouldCompact(contextTokens: number, contextWindow: number, sett
 	return contextTokens > contextWindow - settings.reserveTokens;
 }

+const ESTIMATED_IMAGE_CHARS = 4800;
+
+function estimateTextAndImageContentChars(content: string | Array<{ type: string; text?: string }>): number {
+	if (typeof content === "string") {
+		return content.length;
+	}
+
+	let chars = 0;
+	for (const block of content) {
+		if (block.type === "text" && block.text) {
+			chars += block.text.length;
+		} else if (block.type === "image") {
+			chars += ESTIMATED_IMAGE_CHARS;
+		}
+	}
+	return chars;
+}
+
 /** Estimate token count for one message using a conservative character heuristic. */
 export function estimateTokens(message: AgentMessage): number {
 	let chars = 0;

 	switch (message.role) {
 		case "user": {
-			const content = (message as { content: string | Array<{ type: string; text?: string }> }).content;
-			if (typeof content === "string") {
-				chars = content.length;
-			} else if (Array.isArray(content)) {
-				for (const block of content) {
-					if (block.type === "text" && block.text) {
-						chars += block.text.length;
-					}
-				}
-			}
+			chars = estimateTextAndImageContentChars(
+				(message as { content: string | Array<{ type: string; text?: string }> }).content,
+			);
 			return Math.ceil(chars / 4);
 		}
 		case "assistant": {
@@ -231,18 +248,7 @@ export function estimateTokens(message: AgentMessage): number {
 		}
 		case "custom":
 		case "toolResult": {
-			if (typeof message.content === "string") {
-				chars = message.content.length;
-			} else {
-				for (const block of message.content) {
-					if (block.type === "text" && block.text) {
-						chars += block.text.length;
-					}
-					if (block.type === "image") {
-						chars += 4800;
-					}
-				}
-			}
+			chars = estimateTextAndImageContentChars(message.content);
 			return Math.ceil(chars / 4);
 		}
 		case "bashExecution": {
@@ -281,6 +287,7 @@ function findValidCutPoints(entries: SessionTreeEntry[], startIndex: number, end
 			}
 			case "thinking_level_change":
 			case "model_change":
+			case "active_tools_change":
 			case "compaction":
 			case "branch_summary":
 			case "custom":
@@ -375,7 +382,7 @@ export function findCutPoint(
 	};
 }

-export const SUMMARIZATION_SYSTEM_PROMPT = `You are a context summarization assistant. Your task is to read a conversation between a user and an AI coding assistant, then produce a structured summary following the exact format specified.
+export const SUMMARIZATION_SYSTEM_PROMPT = `You are a context summarization assistant. Your task is to read a conversation between a user and an AI assistant, then produce a structured summary following the exact format specified.

 Do NOT continue the conversation. Do NOT respond to any questions in the conversation. ONLY output the structured summary.`;

--- a/packages/agent/src/harness/compaction/utils.ts
+++ b/packages/agent/src/harness/compaction/utils.ts
@@ -1,4 +1,4 @@
-import type { Message } from "@earendil-works/pi-ai";
+import type { Message } from "@earendil-works/pi-ai/base";
 import type { AgentMessage } from "../../types.ts";

 /** File paths touched by a session branch or compaction range. */
--- a/packages/agent/src/harness/env/nodejs.ts
+++ b/packages/agent/src/harness/env/nodejs.ts
@@ -144,12 +144,25 @@ async function findBashOnPath(): Promise<string | null> {
 	return firstMatch && (await pathExists(firstMatch)) ? firstMatch : null;
 }

-async function getShellConfig(
-	customShellPath?: string,
-): Promise<Result<{ shell: string; args: string[] }, ExecutionError>> {
+interface ShellConfig {
+	shell: string;
+	args: string[];
+	commandTransport?: "argv" | "stdin";
+}
+
+function isLegacyWslBashPath(path: string): boolean {
+	const normalized = path.replace(/\//g, "\\").toLowerCase();
+	return /^[a-z]:\\windows\\(?:system32|sysnative)\\bash\.exe$/.test(normalized);
+}
+
+function getBashShellConfig(shell: string): ShellConfig {
+	return isLegacyWslBashPath(shell) ? { shell, args: ["-s"], commandTransport: "stdin" } : { shell, args: ["-c"] };
+}
+
+async function getShellConfig(customShellPath?: string): Promise<Result<ShellConfig, ExecutionError>> {
 	if (customShellPath) {
 		if (await pathExists(customShellPath)) {
-			return ok({ shell: customShellPath, args: ["-c"] });
+			return ok(getBashShellConfig(customShellPath));
 		}
 		return err(new ExecutionError("shell_unavailable", `Custom shell path not found: ${customShellPath}`));
 	}
@@ -161,22 +174,22 @@ async function getShellConfig(
 		if (programFilesX86) candidates.push(`${programFilesX86}\\Git\\bin\\bash.exe`);
 		for (const candidate of candidates) {
 			if (await pathExists(candidate)) {
-				return ok({ shell: candidate, args: ["-c"] });
+				return ok(getBashShellConfig(candidate));
 			}
 		}
 		const bashOnPath = await findBashOnPath();
 		if (bashOnPath) {
-			return ok({ shell: bashOnPath, args: ["-c"] });
+			return ok(getBashShellConfig(bashOnPath));
 		}
 		return err(new ExecutionError("shell_unavailable", "No bash shell found"));
 	}

 	if (await pathExists("/bin/bash")) {
-		return ok({ shell: "/bin/bash", args: ["-c"] });
+		return ok(getBashShellConfig("/bin/bash"));
 	}
 	const bashOnPath = await findBashOnPath();
 	if (bashOnPath) {
-		return ok({ shell: bashOnPath, args: ["-c"] });
+		return ok(getBashShellConfig(bashOnPath));
 	}
 	return ok({ shell: "sh", args: ["-c"] });
 }
@@ -274,13 +287,22 @@ export class NodeExecutionEnv implements ExecutionEnv {
 			};

 			try {
-				child = spawn(shellConfig.value.shell, [...shellConfig.value.args, command], {
-					cwd,
-					detached: process.platform !== "win32",
-					env: getShellEnv(this.shellEnv, options?.env),
-					stdio: ["ignore", "pipe", "pipe"],
-					windowsHide: true,
-				});
+				const commandFromStdin = shellConfig.value.commandTransport === "stdin";
+				child = spawn(
+					shellConfig.value.shell,
+					commandFromStdin ? shellConfig.value.args : [...shellConfig.value.args, command],
+					{
+						cwd,
+						detached: process.platform !== "win32",
+						env: getShellEnv(this.shellEnv, options?.env),
+						stdio: [commandFromStdin ? "pipe" : "ignore", "pipe", "pipe"],
+						windowsHide: true,
+					},
+				);
+				if (commandFromStdin) {
+					child.stdin?.on("error", () => {});
+					child.stdin?.end(command);
+				}
 			} catch (error) {
 				const cause = toError(error);
 				settle(err(new ExecutionError("spawn_error", cause.message, cause)));
--- a/packages/agent/src/harness/messages.ts
+++ b/packages/agent/src/harness/messages.ts
@@ -1,4 +1,4 @@
-import type { ImageContent, Message, TextContent } from "@earendil-works/pi-ai";
+import type { ImageContent, Message, TextContent } from "@earendil-works/pi-ai/base";
 import type { AgentMessage } from "../types.ts";

 export const COMPACTION_SUMMARY_PREFIX = `The conversation history before this point was compacted into the following summary:
--- a/packages/agent/src/harness/session/session.ts
+++ b/packages/agent/src/harness/session/session.ts
@@ -1,7 +1,8 @@
-import type { ImageContent, TextContent } from "@earendil-works/pi-ai";
+import type { ImageContent, TextContent } from "@earendil-works/pi-ai/base";
 import type { AgentMessage } from "../../types.ts";
 import { createBranchSummaryMessage, createCompactionSummaryMessage, createCustomMessage } from "../messages.ts";
 import type {
+	ActiveToolsChangeEntry,
 	BranchSummaryEntry,
 	CompactionEntry,
 	CustomEntry,
@@ -21,6 +22,7 @@ import { SessionError } from "../types.ts";
 export function buildSessionContext(pathEntries: SessionTreeEntry[]): SessionContext {
 	let thinkingLevel = "off";
 	let model: { provider: string; modelId: string } | null = null;
+	let activeToolNames: string[] | null = null;
 	let compaction: CompactionEntry | null = null;

 	for (const entry of pathEntries) {
@@ -30,6 +32,8 @@ export function buildSessionContext(pathEntries: SessionTreeEntry[]): SessionCon
 			model = { provider: entry.provider, modelId: entry.modelId };
 		} else if (entry.type === "message" && entry.message.role === "assistant") {
 			model = { provider: entry.message.provider, modelId: entry.message.model };
+		} else if (entry.type === "active_tools_change") {
+			activeToolNames = [...entry.activeToolNames];
 		} else if (entry.type === "compaction") {
 			compaction = entry;
 		}
@@ -72,7 +76,7 @@ export function buildSessionContext(pathEntries: SessionTreeEntry[]): SessionCon
 		}
 	}

-	return { messages, thinkingLevel, model };
+	return { messages, thinkingLevel, model, activeToolNames };
 }

 export class Session<TMetadata extends SessionMetadata = SessionMetadata> {
@@ -156,6 +160,16 @@ export class Session<TMetadata extends SessionMetadata = SessionMetadata> {
 		} satisfies ModelChangeEntry);
 	}

+	async appendActiveToolsChange(activeToolNames: string[]): Promise<string> {
+		return this.appendTypedEntry({
+			type: "active_tools_change",
+			id: await this.storage.createEntryId(),
+			parentId: await this.storage.getLeafId(),
+			timestamp: new Date().toISOString(),
+			activeToolNames: [...activeToolNames],
+		} satisfies ActiveToolsChangeEntry);
+	}
+
 	async appendCompaction<T = unknown>(
 		summary: string,
 		firstKeptEntryId: string,
--- a/packages/agent/src/harness/types.ts
+++ b/packages/agent/src/harness/types.ts
@@ -1,5 +1,5 @@
-import type { ImageContent, Model, SimpleStreamOptions, TextContent, Transport } from "@earendil-works/pi-ai";
-import type { AgentEvent, AgentMessage, AgentTool, QueueMode, ThinkingLevel } from "../index.ts";
+import type { ImageContent, Model, SimpleStreamOptions, TextContent, Transport } from "@earendil-works/pi-ai/base";
+import type { AgentEvent, AgentMessage, AgentTool, QueueMode, ThinkingLevel } from "../types.ts";
 import type { Session } from "./session/session.ts";

 /** Result of a fallible operation. Expected failures are returned as `ok: false` instead of thrown. */
@@ -354,6 +354,11 @@ export interface ModelChangeEntry extends SessionTreeEntryBase {
 	modelId: string;
 }

+export interface ActiveToolsChangeEntry extends SessionTreeEntryBase {
+	type: "active_tools_change";
+	activeToolNames: string[];
+}
+
 export interface CompactionEntry<T = unknown> extends SessionTreeEntryBase {
 	type: "compaction";
 	summary: string;
@@ -405,6 +410,7 @@ export type SessionTreeEntry =
 	| MessageEntry
 	| ThinkingLevelChangeEntry
 	| ModelChangeEntry
+	| ActiveToolsChangeEntry
 	| CompactionEntry
 	| BranchSummaryEntry
 	| CustomEntry
@@ -417,6 +423,7 @@ export interface SessionContext {
 	messages: AgentMessage[];
 	thinkingLevel: string;
 	model: { provider: string; modelId: string } | null;
+	activeToolNames: string[] | null;
 }

 export interface SessionMetadata {
@@ -593,19 +600,28 @@ export interface SessionTreeEvent {
 	fromHook?: boolean;
 }

-export interface ModelSelectEvent {
-	type: "model_select";
+export interface ModelUpdateEvent {
+	type: "model_update";
 	model: Model<any>;
 	previousModel: Model<any> | undefined;
 	source: "set" | "restore";
 }

-export interface ThinkingLevelSelectEvent {
-	type: "thinking_level_select";
+export interface ThinkingLevelUpdateEvent {
+	type: "thinking_level_update";
 	level: ThinkingLevel;
 	previousLevel: ThinkingLevel;
 }

+export interface ToolsUpdateEvent {
+	type: "tools_update";
+	toolNames: string[];
+	previousToolNames: string[];
+	activeToolNames: string[];
+	previousActiveToolNames: string[];
+	source: "set" | "restore";
+}
+
 export interface ResourcesUpdateEvent<
 	TSkill extends Skill = Skill,
 	TPromptTemplate extends PromptTemplate = PromptTemplate,
@@ -634,9 +650,10 @@ export type AgentHarnessOwnEvent<
 	| SessionCompactEvent
 	| SessionBeforeTreeEvent
 	| SessionTreeEvent
-	| ModelSelectEvent
-	| ThinkingLevelSelectEvent
-	| ResourcesUpdateEvent<TSkill, TPromptTemplate>;
+	| ModelUpdateEvent
+	| ThinkingLevelUpdateEvent
+	| ResourcesUpdateEvent<TSkill, TPromptTemplate>
+	| ToolsUpdateEvent;

 export type AgentHarnessEvent<TSkill extends Skill = Skill, TPromptTemplate extends PromptTemplate = PromptTemplate> =
 	| AgentEvent
@@ -696,9 +713,10 @@ export type AgentHarnessEventResultMap = {
 	session_compact: undefined;
 	session_before_tree: SessionBeforeTreeResult | undefined;
 	session_tree: undefined;
-	model_select: undefined;
-	thinking_level_select: undefined;
+	model_update: undefined;
+	thinking_level_update: undefined;
 	resources_update: undefined;
+	tools_update: undefined;
 	queue_update: undefined;
 	save_point: undefined;
 	abort: undefined;
--- a/packages/agent/src/index.ts
+++ b/packages/agent/src/index.ts
@@ -1,44 +1,5 @@
-// Core Agent
-export * from "./agent.ts";
-// Loop functions
-export * from "./agent-loop.ts";
-export * from "./harness/agent-harness.ts";
-export {
-	type BranchPreparation,
-	type BranchSummaryDetails,
-	type CollectEntriesResult,
-	collectEntriesForBranchSummary,
-	generateBranchSummary,
-	prepareBranchEntries,
-} from "./harness/compaction/branch-summarization.ts";
-export {
-	calculateContextTokens,
-	compact,
-	DEFAULT_COMPACTION_SETTINGS,
-	estimateContextTokens,
-	estimateTokens,
-	findCutPoint,
-	findTurnStartIndex,
-	generateSummary,
-	getLastAssistantUsage,
-	prepareCompaction,
-	serializeConversation,
-	shouldCompact,
-} from "./harness/compaction/compaction.ts";
-export * from "./harness/messages.ts";
-export * from "./harness/prompt-templates.ts";
-export * from "./harness/session/jsonl-repo.ts";
-export * from "./harness/session/memory-repo.ts";
-export * from "./harness/session/repo-utils.ts";
-export * from "./harness/session/session.ts";
-export { uuidv7 } from "./harness/session/uuid.ts";
-export * from "./harness/skills.ts";
-export * from "./harness/system-prompt.ts";
-// Harness
-export * from "./harness/types.ts";
-export * from "./harness/utils/shell-output.ts";
-export * from "./harness/utils/truncate.ts";
-// Proxy utilities
-export * from "./proxy.ts";
-// Types
-export * from "./types.ts";
+// Import the default pi-ai entrypoint so that all built-in providers register
+// automatically. Unlike "@earendil-works/pi-ai/base" which does not.
+import "@earendil-works/pi-ai";
+
+export * from "./base.ts";
--- a/packages/agent/src/proxy.ts
+++ b/packages/agent/src/proxy.ts
@@ -14,7 +14,7 @@ import {
 	type SimpleStreamOptions,
 	type StopReason,
 	type ToolCall,
-} from "@earendil-works/pi-ai";
+} from "@earendil-works/pi-ai/base";

 // Create stream class matching ProxyMessageEventStream
 class ProxyMessageEventStream extends EventStream<AssistantMessageEvent, AssistantMessage> {
--- a/packages/agent/src/types.ts
+++ b/packages/agent/src/types.ts
@@ -9,7 +9,7 @@ import type {
 	TextContent,
 	Tool,
 	ToolResultMessage,
-} from "@earendil-works/pi-ai";
+} from "@earendil-works/pi-ai/base";
 import type { Static, TSchema } from "typebox";

 /**
@@ -354,7 +354,12 @@ export interface AgentToolResult<T> {
 	terminate?: boolean;
 }

-/** Callback used by tools to stream partial execution updates. */
+/**
+ * Callback used by tools to stream partial execution updates.
+ *
+ * The callback is scoped to the current `execute()` invocation. Calls made after
+ * the tool promise settles are ignored.
+ */
 export type AgentToolUpdateCallback<T = any> = (partialResult: AgentToolResult<T>) => void;

 /** Tool definition used by the agent runtime. */
--- a/packages/agent/test/agent.test.ts
+++ b/packages/agent/test/agent.test.ts
@@ -1,6 +1,7 @@
 import { type AssistantMessage, type AssistantMessageEvent, EventStream, getModel } from "@earendil-works/pi-ai";
+import { Type } from "typebox";
 import { describe, expect, it } from "vitest";
-import { Agent } from "../src/index.ts";
+import { Agent, type AgentEvent, type AgentTool, type AgentToolUpdateCallback } from "../src/index.ts";

 // Mock stream that mimics AssistantMessageEventStream
 class MockAssistantStream extends EventStream<AssistantMessageEvent, AssistantMessage> {
@@ -36,6 +37,28 @@ function createAssistantMessage(text: string): AssistantMessage {
 	};
 }

+type ToolCallContent = Extract<AssistantMessage["content"][number], { type: "toolCall" }>;
+
+function createAssistantToolUseMessage(content: ToolCallContent[]): AssistantMessage {
+	return {
+		role: "assistant",
+		content,
+		api: "openai-responses",
+		provider: "openai",
+		model: "mock",
+		usage: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+			totalTokens: 0,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+		},
+		stopReason: "toolUse",
+		timestamp: Date.now(),
+	};
+}
+
 function createDeferred(): {
 	promise: Promise<void>;
 	resolve: () => void;
@@ -242,6 +265,147 @@ describe("Agent", () => {
 		expect(receivedSignal?.aborted).toBe(true);
 	});

+	it("should ignore tool updates after the tool execution settles", async () => {
+		const toolSchema = Type.Object({});
+		let delayedUpdate: AgentToolUpdateCallback<{ status: string }> | undefined;
+		const events: AgentEvent[] = [];
+		const unhandledRejections: unknown[] = [];
+		const onUnhandledRejection = (error: unknown) => {
+			unhandledRejections.push(error);
+		};
+		const tool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "delayed_tool",
+			label: "Delayed Tool",
+			description: "Captures progress callbacks",
+			parameters: toolSchema,
+			async execute(_toolCallId, _params, _signal, onUpdate) {
+				delayedUpdate = onUpdate;
+				onUpdate?.({
+					content: [{ type: "text", text: "running" }],
+					details: { status: "running" },
+				});
+				return {
+					content: [{ type: "text", text: "ok" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const agent = new Agent({
+			initialState: { tools: [tool] },
+			streamFn: () => {
+				const stream = new MockAssistantStream();
+				queueMicrotask(() => {
+					stream.push({
+						type: "done",
+						reason: "toolUse",
+						message: createAssistantToolUseMessage([
+							{ type: "toolCall", id: "call-1", name: "delayed_tool", arguments: {} },
+						]),
+					});
+				});
+				return stream;
+			},
+		});
+		agent.subscribe((event) => {
+			events.push(event);
+		});
+
+		process.on("unhandledRejection", onUnhandledRejection);
+		try {
+			await agent.prompt("run tool");
+			const eventCountAfterPrompt = events.length;
+
+			delayedUpdate?.({
+				content: [{ type: "text", text: "late" }],
+				details: { status: "late" },
+			});
+			await new Promise((resolve) => setTimeout(resolve, 0));
+
+			expect(events.filter((event) => event.type === "tool_execution_update")).toHaveLength(1);
+			expect(events).toHaveLength(eventCountAfterPrompt);
+			expect(unhandledRejections).toEqual([]);
+		} finally {
+			process.off("unhandledRejection", onUnhandledRejection);
+		}
+	});
+
+	it("should ignore a settled parallel tool update while another tool is still running", async () => {
+		const toolSchema = Type.Object({});
+		const slowStarted = createDeferred();
+		const settledToolEnded = createDeferred();
+		const releaseSlow = createDeferred();
+		let settledToolUpdate: AgentToolUpdateCallback<{ status: string }> | undefined;
+		const events: AgentEvent[] = [];
+		const settledTool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "settled_tool",
+			label: "Settled Tool",
+			description: "Captures progress callbacks",
+			parameters: toolSchema,
+			async execute(_toolCallId, _params, _signal, onUpdate) {
+				settledToolUpdate = onUpdate;
+				return {
+					content: [{ type: "text", text: "done" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const slowTool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "slow_tool",
+			label: "Slow Tool",
+			description: "Keeps the agent run active",
+			parameters: toolSchema,
+			async execute() {
+				slowStarted.resolve();
+				await releaseSlow.promise;
+				return {
+					content: [{ type: "text", text: "done" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const agent = new Agent({
+			initialState: { tools: [settledTool, slowTool] },
+			streamFn: () => {
+				const stream = new MockAssistantStream();
+				queueMicrotask(() => {
+					stream.push({
+						type: "done",
+						reason: "toolUse",
+						message: createAssistantToolUseMessage([
+							{ type: "toolCall", id: "call-1", name: "settled_tool", arguments: {} },
+							{ type: "toolCall", id: "call-2", name: "slow_tool", arguments: {} },
+						]),
+					});
+				});
+				return stream;
+			},
+		});
+		agent.subscribe((event) => {
+			events.push(event);
+			if (event.type === "tool_execution_end" && event.toolCallId === "call-1") {
+				settledToolEnded.resolve();
+			}
+		});
+
+		const promptPromise = agent.prompt("run tools");
+		await Promise.all([slowStarted.promise, settledToolEnded.promise]);
+		const eventCountBeforeLateUpdate = events.length;
+
+		settledToolUpdate?.({
+			content: [{ type: "text", text: "late" }],
+			details: { status: "late" },
+		});
+		await new Promise((resolve) => setTimeout(resolve, 0));
+		expect(events).toHaveLength(eventCountBeforeLateUpdate);
+
+		releaseSlow.resolve();
+		await promptPromise;
+		expect(events.filter((event) => event.type === "tool_execution_update")).toHaveLength(0);
+	});
+
 	it("should update state with mutators", () => {
 		const agent = new Agent();

--- a/packages/agent/test/harness/agent-harness.test.ts
+++ b/packages/agent/test/harness/agent-harness.test.ts
@@ -17,10 +17,6 @@ interface AppPromptTemplate extends PromptTemplate {
 	source: "project" | "user";
 }

-interface AppTool extends AgentTool {
-	source: "builtin" | "extension";
-}
-
 const registrations: Array<{ unregister(): void }> = [];

 function textFromUserMessages(messages: Array<{ role: string; content: unknown }>): string[] {
@@ -458,11 +454,111 @@ describe("AgentHarness", () => {
 		});
 	});

+	it("preserves app tool types for getters and update events", async () => {
+		const session = new Session(new InMemorySessionStorage());
+		const env = new NodeExecutionEnv({ cwd: process.cwd() });
+		const model = getModel("anthropic", "claude-sonnet-4-5");
+		type AppTool = AgentTool<typeof calculateTool.parameters, undefined> & { source: "builtin" | "extension" };
+		const inspectTool: AppTool = { ...calculateTool, name: "inspect", source: "builtin" };
+		const searchTool: AppTool = { ...calculateTool, name: "search", source: "extension" };
+		const harness = new AgentHarness<AppSkill, AppPromptTemplate, AppTool>({
+			env,
+			session,
+			model,
+			tools: [inspectTool, searchTool],
+			activeToolNames: ["inspect"],
+		});
+		const updates: Array<{
+			toolNames: string[];
+			previousToolNames: string[];
+			activeToolNames: string[];
+			previousActiveToolNames: string[];
+			source: "set" | "restore";
+		}> = [];
+		harness.subscribe((event) => {
+			if (event.type === "tools_update") {
+				updates.push({
+					toolNames: event.toolNames,
+					previousToolNames: event.previousToolNames,
+					activeToolNames: event.activeToolNames,
+					previousActiveToolNames: event.previousActiveToolNames,
+					source: event.source,
+				});
+				expect(harness.getActiveTools().map((tool) => tool.name)).toEqual(event.activeToolNames);
+			}
+		});
+
+		const tools = harness.getTools();
+		const activeTools = harness.getActiveTools();
+		tools.pop();
+		activeTools.pop();
+		expect(harness.getTools().map((tool) => tool.name)).toEqual(["inspect", "search"]);
+		expect(harness.getActiveTools().map((tool) => tool.source)).toEqual(["builtin"]);
+
+		await harness.setActiveTools(["search"]);
+		await harness.setTools([searchTool], ["search"]);
+		await expect(harness.setActiveTools(["missing"])).rejects.toMatchObject({ code: "invalid_argument" });
+		await expect(harness.setActiveTools(["search", "search"])).rejects.toMatchObject({ code: "invalid_argument" });
+		await expect(harness.setTools([inspectTool])).rejects.toMatchObject({ code: "invalid_argument" });
+		await expect(harness.setTools([inspectTool, inspectTool], ["inspect"])).rejects.toMatchObject({
+			code: "invalid_argument",
+		});
+
+		expect(updates).toEqual([
+			{
+				toolNames: ["inspect", "search"],
+				previousToolNames: ["inspect", "search"],
+				activeToolNames: ["search"],
+				previousActiveToolNames: ["inspect"],
+				source: "set",
+			},
+			{
+				toolNames: ["search"],
+				previousToolNames: ["inspect", "search"],
+				activeToolNames: ["search"],
+				previousActiveToolNames: ["search"],
+				source: "set",
+			},
+		]);
+		expect(harness.getTools().map((tool) => tool.source)).toEqual(["extension"]);
+		expect(harness.getActiveTools().map((tool) => tool.name)).toEqual(["search"]);
+		expect((await session.buildContext()).activeToolNames).toEqual(["search"]);
+	});
+
+	it("validates constructor tool names", () => {
+		const session = new Session(new InMemorySessionStorage());
+		const env = new NodeExecutionEnv({ cwd: process.cwd() });
+		const model = getModel("anthropic", "claude-sonnet-4-5");
+		expect(
+			() => new AgentHarness({ env, session, model, tools: [calculateTool], activeToolNames: ["missing"] }),
+		).toThrow(/Unknown tool/);
+		expect(
+			() =>
+				new AgentHarness({
+					env,
+					session,
+					model,
+					tools: [calculateTool, calculateTool],
+					activeToolNames: [calculateTool.name],
+				}),
+		).toThrow(/Duplicate tool/);
+		expect(
+			() =>
+				new AgentHarness({
+					env,
+					session,
+					model,
+					tools: [calculateTool],
+					activeToolNames: [calculateTool.name, calculateTool.name],
+				}),
+		).toThrow(/Duplicate active tool/);
+	});
+
 	it("preserves app resource types for getters and update events", async () => {
 		const session = new Session(new InMemorySessionStorage());
 		const env = new NodeExecutionEnv({ cwd: process.cwd() });
 		const model = getModel("anthropic", "claude-sonnet-4-5");
-		const harness = new AgentHarness<AppSkill, AppPromptTemplate, AppTool>({ env, session, model });
+		const harness = new AgentHarness<AppSkill, AppPromptTemplate, AgentTool>({ env, session, model });
 		const skill: AppSkill = {
 			name: "inspect",
 			description: "Inspect things",
--- a/packages/agent/test/harness/nodejs-env.test.ts
+++ b/packages/agent/test/harness/nodejs-env.test.ts
@@ -1,5 +1,5 @@
 import { access, chmod, realpath, symlink } from "node:fs/promises";
-import { join } from "node:path";
+import { delimiter, join } from "node:path";
 import { afterEach, describe, expect, it } from "vitest";
 import { NodeExecutionEnv } from "../../src/harness/env/nodejs.ts";
 import { FileError, getOrThrow } from "../../src/harness/types.ts";
@@ -201,6 +201,39 @@ describe("NodeExecutionEnv", () => {
 		expect(result).toEqual({ stdout: `${await realpath(root)}:ok`, stderr: "", exitCode: 0 });
 	});

+	it("uses stdin command transport for legacy WSL bash paths", async () => {
+		if (process.platform === "win32") return;
+		const root = createTempDir();
+		const shellPath = "C:\\Windows\\System32\\bash.exe";
+		const env = new NodeExecutionEnv({ cwd: root });
+		getOrThrow(await env.writeFile(shellPath, '#!/bin/sh\nprintf \'args:%s\\n\' "$*" >&2\nexec /bin/bash "$@"\n'));
+		await chmod(join(root, shellPath), 0o755);
+
+		const originalCwd = process.cwd();
+		const originalPath = process.env.PATH;
+		const platformDescriptor = Object.getOwnPropertyDescriptor(process, "platform");
+		try {
+			process.chdir(root);
+			process.env.PATH = `${root}${delimiter}${originalPath ?? ""}`;
+			Object.defineProperty(process, "platform", {
+				configurable: true,
+				value: "win32",
+			});
+
+			const wslEnv = new NodeExecutionEnv({ cwd: root, shellPath });
+			const nameExpansion = "$" + "{name}";
+			const result = getOrThrow(await wslEnv.exec(`name='World'; echo "Hello, ${nameExpansion}!"`));
+
+			expect(result).toEqual({ stdout: "Hello, World!\n", stderr: "args:-s\n", exitCode: 0 });
+		} finally {
+			process.chdir(originalCwd);
+			process.env.PATH = originalPath;
+			if (platformDescriptor) {
+				Object.defineProperty(process, "platform", platformDescriptor);
+			}
+		}
+	});
+
 	it("streams stdout and stderr chunks", async () => {
 		const root = createTempDir();
 		const env = new NodeExecutionEnv({ cwd: root });
--- a/packages/agent/tsconfig.build.json
+++ b/packages/agent/tsconfig.build.json
@@ -2,6 +2,10 @@
 	"extends": "../../tsconfig.base.json",
 	"compilerOptions": {
 		"outDir": "./dist",
+		"paths": {
+			"@earendil-works/pi-ai": ["../ai/dist/index.d.ts"],
+			"@earendil-works/pi-ai/*": ["../ai/dist/*.d.ts", "../ai/dist/providers/*.d.ts"]
+		},
 		"rootDir": "./src"
 	},
 	"include": ["src/**/*.ts"],
--- a/packages/agent/vitest.config.ts
+++ b/packages/agent/vitest.config.ts
@@ -1,6 +1,12 @@
 import { defineConfig } from "vitest/config";

 export default defineConfig({
+	resolve: {
+		alias: {
+			"@earendil-works/pi-ai/base": new URL("../ai/src/base.ts", import.meta.url).pathname,
+			"@earendil-works/pi-ai": new URL("../ai/src/index.ts", import.meta.url).pathname,
+		},
+	},
 	test: {
 		globals: true,
 		environment: "node",
--- a/packages/agent/vitest.harness.config.ts
+++ b/packages/agent/vitest.harness.config.ts
@@ -1,6 +1,12 @@
 import { defineConfig } from "vitest/config";

 export default defineConfig({
+	resolve: {
+		alias: {
+			"@earendil-works/pi-ai/base": new URL("../ai/src/base.ts", import.meta.url).pathname,
+			"@earendil-works/pi-ai": new URL("../ai/src/index.ts", import.meta.url).pathname,
+		},
+	},
 	test: {
 		globals: true,
 		environment: "node",