// Creates or updates Claude Code skills through interactive dialog, then evaluates effectiveness by parallel execution comparison. Use when creating new skills, updating existing skills, or evaluating skill quality.
| name | recipe-eval-skill |
| description | Creates or updates Claude Code skills through interactive dialog, then evaluates effectiveness by parallel execution comparison. Use when creating new skills, updating existing skills, or evaluating skill quality. |
| disable-model-invocation | true |
Context: Skill authoring (Phase A) followed by blind A/B evaluation (Phase B)
Mode: $ARGUMENTS
Core Identity: "I am not a worker. I am an orchestrator."
Execution Method:
claude -p)Orchestrator invokes sub-agents via Agent tool and scripts via Bash, passes structured data between them.
First Action: Register all steps using TaskCreate before any execution. Phase A steps are defined in the mode-specific reference (create.md or update.md). Phase B steps are defined in eval.md. Update status using TaskUpdate upon each step completion.
Determine mode from $ARGUMENTS:
| Mode | Criteria |
|---|---|
| Creation | "create", new skill request, no existing skill referenced |
| Update | "improve", "update", existing skill name or path mentioned |
| Unspecified | $ARGUMENTS is empty or ambiguous |
Phase A (Skill Authoring): Create or modify skill content through dialog. Ends with user-approved skill file. Phase B (Evaluation): Measure skill effectiveness through blind execution comparison. Does not modify skill content.
Responsibility Boundary: This skill completes with the combined evaluation report and ship/revise/reject recommendation.
Read the mode-specific reference and execute:
Phase A ends with: user-approved skill content (new or modified).
Before starting Phase B, confirm these data are available in context. Phase B cannot proceed without them:
| Data | Source | Required |
|---|---|---|
| Skill name | Phase A dialog | Always |
| Source skill directory | Phase A file write | Always |
| User phrases | Phase A Round 3 (create) / Round 2 (update) | Always |
| Trigger scenarios | Phase A Round 3 (create) / Round 1-2 (update) | Always |
| Original SKILL.md content | Phase A Step 6 (update mode only) | Update mode |
If user phrases are missing, ask the user before proceeding: "What phrases does your team use when requesting work that this skill covers?"
Read references/eval.md and execute the evaluation protocol. Pass the handoff data above as context.
Phase B consists of:
Present combined results to user:
| Scenario | Behavior |
|---|---|
| User cancels during Phase A | Stop. No eval needed. |
| Grade C after 2 iterations | Present content with issues. User decides: accept/revise/abort. |
| One executor fails in Phase B | Continue with partial comparison. |
| Both executors fail in Phase B | Report failure. Phase A result still valid. |
| Worktree creation fails | Report git error. Phase A result still valid. |
claude CLI available in PATHAnalyzes and optimizes prompts using BP-001~008 patterns and 3-step flow (detect, optimize, balance). Use when "optimize this prompt", "review prompt quality", "analyze prompt issues", or creating/reviewing rashomon skill content.
Project-specific prompt optimization knowledge management. Use when storing or retrieving learned patterns from comparisons. Provides schema, extraction criteria, capacity management, and retention scoring.
Git worktree management for isolated parallel prompt execution. Use when creating isolated environments for prompt comparison or managing worktree lifecycle. Provides creation, cleanup, and orphan detection scripts.
Compares original and optimized prompts by parallel execution in git worktrees. Use when evaluating prompt improvement effects or learning prompt engineering through concrete examples.