mirror of
https://github.com/n8n-io/n8n.git
synced 2026-10-19 07:46:02 +02:00
d3bc90a777
Correlate tool_result events back to their tool_use by id and record every
is_error result as {tool, message} (message truncated to 500 chars; string
and text-block-array contents both handled). The failure list rides next to
toolCalls through the whole spend pipeline: buildToolErrorsPerRun per case
and per-tool error counts in summary.mcpBuild.toolErrors in
eval-results.json, a build_tool_errors count as LangSmith row feedback plus
per-tool totals in experiment metadata, and per-build lists + totals in the
manifest builder's stats sidecar. Timed-out builds keep contributing the
errors their partial stream recorded.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GA23ytEUfHpzkv7meRdVAJ
|
||
|---|---|---|
| .. | ||
| args.ts | ||
| build-cost-report.ts | ||
| build-mcp-manifest.ts | ||
| compare-pairwise.ts | ||
| index.ts | ||
| langtracer-push.ts | ||
| mcp-builder.ts | ||
| pairwise.ts | ||
| report.ts | ||