fix(agents): skip error/aborted assistant messages in transcript repair by lailoo · Pull Request #4844 · openclaw/openclaw

lailoo · 2026-01-30T19:02:31Z

Summary

When an assistant response is interrupted mid-tool-call (stopReason: error or aborted), repairToolUseResultPairing() was still extracting tool calls and inserting synthetic tool_result entries. This caused orphan tool_result blocks after transformMessages() dropped the errored assistant, leading to permanent session corruption with unexpected tool_use_id API errors.

Changes

Modified repairToolUseResultPairing() in src/agents/session-transcript-repair.ts to skip tool call extraction for stopReason === "error" || "aborted" assistant messages
Added 4 new unit tests covering error/aborted scenarios

Root Cause

Two safety mechanisms interacted badly:

session-tool-result-guard inserts synthetic toolResult for orphaned tool calls
transformMessages() drops error/aborted assistant messages but not their corresponding toolResult

This created orphaned tool_result entries that referenced non-existent tool_use blocks.

Testing

All 8 tests in session-transcript-repair.test.ts pass
Related tests in session-tool-result-guard.test.ts and pi-embedded-runner.guard.test.ts pass

Fixes

Terminated assistant with toolCall causes infinite 'unexpected tool_use_id' API rejection loop #4600 - Terminated assistant with toolCall causes infinite rejection loop
fix(agents): handle stopReason 'aborted' in transcript repair to prevent session corruption #4597 - handle stopReason 'aborted' in transcript repair
Session corruption when assistant tool call is aborted mid-stream #4475 - Session corruption when tool call aborted mid-stream
Transcript repair (repairToolUseResultPairing) failed to catch orphaned tool_use block #4814 - repairToolUseResultPairing failed to catch orphaned tool_use
Transcript repair (repairToolUseResultPairing) failed to catch orphaned tool_use block #4815 - repairToolUseResultPairing failed to catch orphaned tool_use

Greptile Overview

Greptile Summary

This PR updates repairToolUseResultPairing() to avoid extracting tool calls (and thus avoid inserting synthetic toolResult entries) from assistant messages with stopReason: "error" or "aborted". This prevents creating orphan toolResult blocks later when other sanitization (transformMessages()) drops those assistant turns but leaves tool results behind.

Unit tests were added to ensure error/aborted assistant messages do not trigger synthetic toolResult insertion, while normal tool-using assistants are still repaired as before.

Confidence Score: 4/5

This PR is likely safe to merge and directly addresses a real session-corruption failure mode.
The change is small, localized, and covered by new unit tests for the targeted stopReason cases. Main remaining uncertainty is behavioral: skipping errored/aborted assistants could drop legitimate tool results that followed those turns, which may or may not be desired depending on how transcripts are recorded in practice.
src/agents/session-transcript-repair.ts

When an assistant response is interrupted mid-tool-call (stopReason: error or aborted), repairToolUseResultPairing() was still extracting tool calls and inserting synthetic tool_result entries. This caused orphan tool_result blocks after transformMessages() dropped the errored assistant, leading to permanent session corruption with 'unexpected tool_use_id' API errors. This fix skips tool call extraction for error/aborted assistant messages, preventing the creation of orphan tool_result entries. Fixes openclaw#4600, openclaw#4597, openclaw#4475, openclaw#4814, openclaw#4815

greptile-apps

_{2 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-02-03T02:25:13Z

src/agents/session-transcript-repair.ts

+    // Skip error/aborted assistant messages - do not extract tool calls from them
+    // as they may contain incomplete/invalid tool_use blocks that would cause
+    // orphan tool_result entries after transformMessages() drops the errored assistant
+    const stopReason = (assistant as { stopReason?: string }).stopReason;


[P1] Skipping errored/aborted assistants can leave their existing tool results as orphans.

This change avoids creating synthetic toolResults for stopReason === "error" | "aborted", but if the transcript already contains real tool results for those tool calls (e.g., assistant streamed toolCall then later got marked error, yet the tool executed and logged a toolResult), the next loop iterations will hit the role !== "assistant" branch and drop those toolResults as “orphan”. That can discard valid tool outputs and potentially change downstream behavior (e.g., losing tool error context).

If that scenario is expected, it likely needs a test and/or a more nuanced rule (e.g., treat errored assistants with complete tool calls differently, or only skip when the assistant content is known-incomplete).

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/session-transcript-repair.ts Line: 120:123 Comment: [P1] Skipping errored/aborted assistants can leave their *existing* tool results as orphans. This change avoids creating synthetic `toolResult`s for `stopReason === "error" | "aborted"`, but if the transcript already contains real tool results for those tool calls (e.g., assistant streamed `toolCall` then later got marked `error`, yet the tool executed and logged a `toolResult`), the next loop iterations will hit the `role !== "assistant"` branch and drop those `toolResult`s as “orphan”. That can discard valid tool outputs and potentially change downstream behavior (e.g., losing tool error context). If that scenario is expected, it likely needs a test and/or a more nuanced rule (e.g., treat errored assistants with complete tool calls differently, or only skip when the assistant content is known-incomplete). How can I resolve this? If you propose a fix, please make it concise.

greptile-apps · 2026-02-03T02:25:14Z

src/agents/session-transcript-repair.test.ts

+  it("skips tool call extraction for error assistant messages", () => {
+    const input = [
+      {
+        role: "assistant",
+        stopReason: "error",
+        errorMessage: "Request terminated",


[P2] New tests only assert “no synthetic toolResult inserted”, but don’t cover the case where a real toolResult already exists.

Given the production bug involves interactions with transformMessages() dropping errored assistants, it would be useful to add a test where an assistant(stopReason:error) with a tool call is followed by an actual matching toolResult entry, and verify the intended behavior (drop it? keep it? move it?). Without that, the behavior change mentioned in session-transcript-repair.ts can regress silently.

Prompt To Fix With AI

This is a comment left during a code review. Path: src/agents/session-transcript-repair.test.ts Line: 113:118 Comment: [P2] New tests only assert “no synthetic toolResult inserted”, but don’t cover the case where a *real* toolResult already exists. Given the production bug involves interactions with `transformMessages()` dropping errored assistants, it would be useful to add a test where an `assistant(stopReason:error)` with a tool call is followed by an actual matching `toolResult` entry, and verify the intended behavior (drop it? keep it? move it?). Without that, the behavior change mentioned in `session-transcript-repair.ts` can regress silently. How can I resolve this? If you propose a fix, please make it concise.

Glucksberg · 2026-02-08T12:02:07Z

👋 This appears to duplicate PR #4476, which implements the same fix. Consider consolidating or closing one to avoid duplicate effort.

steipete · 2026-02-24T01:19:32Z

AI-assisted stale triage closure (2026-02-24).

Closing this PR because the fix is already in main.

Why:

This stopReason error/aborted transcript-skip behavior was merged via fix(agents): skip tool extraction for aborted/errored assistant messages #4598 and is now present in current repair logic.

This is AI-closed housekeeping, not a rejection of the issue.

If you still see orphan tool-result behavior after error/abort paths, open a fresh focused PR from latest main with a repro.

openclaw-barnacle bot added the agents Agent runtime and tooling label Jan 30, 2026

This was referenced Feb 2, 2026

fix(session): strip malformed tool_use blocks to prevent session corruption #5557

Closed

fix(agents): skip extracting tool calls from errored assistant turns #1859

Closed

greptile-apps bot reviewed Feb 3, 2026

View reviewed changes

quotentiroler mentioned this pull request Feb 6, 2026

fix(agents): skip tool extraction for aborted/errored assistant messages #4598

Merged

This was referenced Feb 8, 2026

fix: skip synthetic tool results for aborted/errored messages #11729

Closed

Transcript repair creates orphaned tool_result for aborted tool calls #6788

Closed

sebslight mentioned this pull request Feb 13, 2026

fix: skip synthetic tool_result for aborted assistant messages #6831

Closed

3 tasks

thewilloftheshadow force-pushed the main branch from bfc1ccb to f92900f Compare February 15, 2026 18:46

sebslight mentioned this pull request Feb 23, 2026

Agents: skip errored tool calls during pairing #7525

Closed

steipete closed this Feb 24, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(agents): skip error/aborted assistant messages in transcript repair#4844

fix(agents): skip error/aborted assistant messages in transcript repair#4844
lailoo wants to merge 1 commit intoopenclaw:mainfrom
lailoo:fix/session-transcript-repair-error-handling

lailoo commented Jan 30, 2026 •

edited by greptile-apps bot

Loading

Uh oh!

greptile-apps bot left a comment

Uh oh!

greptile-apps bot Feb 3, 2026

Uh oh!

greptile-apps bot Feb 3, 2026

Uh oh!

Glucksberg commented Feb 8, 2026

Uh oh!

steipete commented Feb 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

lailoo commented Jan 30, 2026 • edited by greptile-apps bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Root Cause

Testing

Fixes

Greptile Overview

Greptile Summary

Confidence Score: 4/5

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 3, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 3, 2026

Choose a reason for hiding this comment

Uh oh!

Glucksberg commented Feb 8, 2026

Uh oh!

steipete commented Feb 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lailoo commented Jan 30, 2026 •

edited by greptile-apps bot

Loading