fix(coding-agent): resolve waitForRetry() race when auto-retry produces tool calls (#2440)

When auto-retry fires after a retryable error (e.g. overloaded_error) and the
retry response includes tool_use, session.prompt() returned prematurely because
_resolveRetry() was called on the first successful message_end — while the
agent loop was still executing tools via the fire-and-forget agent.continue().

This caused callers to observe isStreaming=true after prompt() returned, and
follow-up session.prompt() calls threw 'Agent is already processing'. The
tool execution results were silently lost.

Fix: move _resolveRetry() from the message_end handler to the agent_end
handler. The _retryAttempt counter reset stays on message_end (preventing
accumulation across LLM calls within a turn), but the promise that unblocks
waitForRetry() now only resolves when the full agent loop completes.
This commit is contained in:
Petr Baudis
2026-03-20 00:33:17 +01:00
committed by GitHub
parent 970774ec3c
commit 8a0529ed9d
2 changed files with 91 additions and 2 deletions

View File

@@ -490,7 +490,6 @@ export class AgentSession {
attempt: this._retryAttempt,
});
this._retryAttempt = 0;
this._resolveRetry();
}
}
}
@@ -506,6 +505,7 @@ export class AgentSession {
if (didRetry) return; // Retry was initiated, don't proceed to compaction
}
this._resolveRetry();
await this._checkCompaction(msg);
}
}