fix: validate and fix tool_result IDs before API requests #9952
Conversation
Fixes ApiError: Expected toolResult blocks for mismatched IDs ## Problem Tool_result blocks sometimes had incorrect tool_use_id values, causing the Anthropic API to reject requests with errors like: "Expected toolResult blocks at messages.6.content for the following Ids: toolu_ABC, but found: toolu_XYZ" ## Solution Added centralized validation in addToApiConversationHistory() that: - Validates tool_result IDs against tool_use IDs from previous assistant message - Fixes mismatches using position-based matching - Logs warnings for debugging when corrections are made ## Changes - Added src/core/task/validateToolResultIds.ts with validation logic - Added src/core/task/__tests__/validateToolResultIds.spec.ts (12 tests) - Modified src/core/task/Task.ts to call validation - Simplified src/core/assistant-message/presentAssistantMessage.ts by removing redundant source-level validation
dosubot
bot
added
size:XL
hannesrudolph
added
the
Issue/PR - Triage
Review complete. No issues found. This PR adds centralized validation to correct tool_result ID mismatches before API requests. The implementation is well-structured with comprehensive test coverage (12 tests) and handles edge cases gracefully. The position-based matching strategy is appropriate for the stated use cases (race conditions, message editing, resume/delegation scenarios). Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues. |
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
Review complete. No issues found. This commit (11fdfd8) fixes a subtle bug in position-based matching when duplicate tool_use_ids are present:
All previous review comments remain resolved:
Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues. |
|
I think in general it would be nice to fully understand the root cause of this, and a solid repro case would help with that. Short of that, we should considering emitting an event to Posthog every time we detect that a correct is needed. |
- Replace .slice().reverse().find() with direct backwards loop - Avoids creating two intermediate arrays for large conversation histories - Applied to both addToApiConversationHistory and flushPendingToolResultsToHistory
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
- Replace manual backwards loop with findLastIndex in both addToApiConversationHistory and flushPendingToolResultsToHistory - Improves clarity and consistency with existing code patterns
daniel-lxs
force-pushed
the
fix/tool-result-id-validation
branch
from
f9952b0 to
ddf88d8
Compare
Problem
Tool_result blocks sometimes had incorrect tool_use_id values, causing the Anthropic API to reject requests with errors like:
This could happen due to:
Solution
Added centralized validation in
addToApiConversationHistory()that:Changes
src/core/task/validateToolResultIds.tswith centralized validation logicsrc/core/task/__tests__/validateToolResultIds.spec.tswith 12 comprehensive testssrc/core/task/Task.tsto call validation inaddToApiConversationHistory()src/core/assistant-message/presentAssistantMessage.tsby removing redundant source-level validationPerformance
The validation adds negligible overhead (~0.1-0.5ms) compared to API call times (500ms-30s). It only runs on user messages containing tool_result blocks.
Testing
All 4589 tests pass.
Important
Add validation and correction for
tool_resultIDs in user messages to match previous assistanttool_useIDs, with telemetry and tests.validateAndFixToolResultIds()andToolResultIdMismatchErrorinvalidateToolResultIds.tsto validate/correcttool_result.tool_use_idby position against prior assistanttool_useIDs; reports mismatches via telemetry.Task.addToApiConversationHistory()andflushPendingToolResultsToHistory(), run validation on user messages before pushing toapiConversationHistory.validateToolResultIds.spec.tswith comprehensive cases (matching, mismatched, extra/missing items, non-tool content, string content, telemetry).This description was created by
for 11fdfd8. You can customize this summary. It will automatically update as commits are pushed.
Note
Validates and auto-corrects
tool_result.tool_use_idvalues against the previous assistanttool_useblocks before persisting user messages, with telemetry and tests.validateAndFixToolResultIds()andToolResultIdMismatchErrorinsrc/core/task/validateToolResultIds.tsto validate/correcttool_result.tool_use_idby position against prior assistanttool_useIDs; reports mismatches via telemetry.Task.addToApiConversationHistory()andflushPendingToolResultsToHistory(), run validation on user messages before pushing toapiConversationHistory.src/core/task/__tests__/validateToolResultIds.spec.tswith comprehensive cases (matching, mismatched, extra/missing items, non-tool content, string content, telemetry).Written by Cursor Bugbot for commit c07b9f9. This will update automatically on new commits. Configure here.