fix: improve skill trigger routing accuracy across models#1178
Merged
Conversation
Sharpen skill descriptions to reduce cross-model misrouting identified by multi-model eval runs. Adds explicit exclusions to extension-model, extension-driver, and model skills. Enriches report description with differentiating keywords. Strengthens eval system prompt against text-only responses. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extension-model was incorrectly triggering on publishing queries like "Prepare my extension for publishing" and "Publish my model to the registry" after the initial description narrowing. Add explicit exclusion directing these to swamp-extension-publish. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
Blocking Issues
None.
Suggestions
- Pre-existing body line limit:
swamp-extension-model/SKILL.mdis ~509 body lines (529 total minus ~20 lines of frontmatter), slightly exceeding the 500-line body limit in CLAUDE.md. This is pre-existing and not introduced by this PR, but worth noting for a future cleanup pass to split content intoreferences/files.
Summary
Clean, focused PR that sharpens skill frontmatter descriptions to reduce cross-model misrouting. The changes are well-structured:
- Explicit "Do NOT use for X — that is Y" exclusions on
swamp-extension-driver,swamp-extension-model,swamp-model, andswamp-reportprovide clear disambiguation guidance - Differentiating API keywords (
dataRepository,UnifiedDataRepository,dataHandles,MethodReportContext) added toswamp-reportto fix report routing failures "remote execution"trigger correctly removed fromswamp-extension-driverto prevent workflow misrouting- Eval system prompt reinforcement ("A text-only response with no tool call is ALWAYS wrong") is a good nudge for models that default to text responses
- All YAML frontmatter uses valid
>block scalars with correctname/description-only fields - CI passes: lint, test, format, skill review, and skill trigger eval all green
- Eval results show clear improvement (Opus 94.1% → 96.5%, targeted failures fixed)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
swamp-report,swamp-extension-model,swamp-extension-driver, andswamp-modelto reduce cross-model misrouting identified by multi-model eval runsdataRepository,UnifiedDataRepository,dataHandles,MethodReportContext) toswamp-reportso report API queries stop routing to troubleshootingEval Results
Validated against multi-model eval suite (run 1, run 2):
*Gemini rate-limited, 128/202 tests completed.
Original cross-model failures from issue:
Closes #80
Test plan
deno checkpassesdeno lintpassesdeno fmtpassesdeno run compilesucceeds🤖 Generated with Claude Code