| name | yuuzu-ide-skill-eval-feedback |
| description | Evaluate and improve repo-local Yuuzu-IDE testing skills. Use this when creating or updating skills under .codex/skills/, adding evals/evals.json, comparing with-skill and baseline outputs, or collecting feedback on which Yuuzu-IDE test methods are slow, flaky, redundant, unsafe, or missing runtime coverage. |
Yuuzu-IDE Skill Eval Feedback
Use this skill to keep the project testing skills useful instead of letting them become static documentation.
Workflow
- Identify the skill under review and read its
SKILL.md.
- Read that skill's
evals/evals.json.
- Read
references/eval-feedback-rubric.md.
- Validate the evals statically before launching any run:
- JSON parses.
skill_name matches frontmatter.
- prompts are realistic Yuuzu-IDE tasks.
- expected outputs are checkable.
- expectations can reveal bad testing advice, not just keyword presence.
- Only run with-skill/baseline eval agents when the user explicitly asks for eval execution or delegated skill benchmarking.
- When eval outputs exist, grade them against expectations and add feedback about the testing method itself: speed, reliability, safety, evidence quality, and missing coverage.
Feedback output
Use this shape:
Skill:
Eval coverage:
Pass/fail signals:
Test-method issues:
Optimization proposals:
Run next:
Optimization proposals should be actionable, for example:
- Replace broad
bun test first steps with a focused adjacent test.
- Add a runtime smoke only for visual/packaged-app behavior.
- Split a flaky UI smoke into one deterministic component test plus one smaller real-app probe.
- Add a Rust workspace-boundary test when a frontend test only mocks IPC.