prompt-iteration
Use when iteratively improving agent prompts through automated LLM-as-Judge evaluation. Runs eval→fix→commit loop with circuit breakers.
npx skills add majiayu000/claude-skill-registry --skill prompt-iteration --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
# Prompt Iteration Skill Iteratively improve teaching agent prompts using automated evaluation with local Qwen 2.5-7B as judge. ## When to Use - After identifying functional issues with the agent - When prompt changes need validation before deployment - For systematic prompt engineering with measurable improvement ## Pre-flight Checks Before starting, verify: 1. **vLLM is running**: ```bash curl http://localhost:8004/v1/models ``` 2. **Golden dataset exists**: ```bash ls tests/golden/ ``` 3. **Create iteration branch**: ```bash git checkout -b eval/prompt-iteration-$(date +%Y%m%d-%H%M) ``` ## Guardrails (USER-DEFINED) | Guardrail | Value | |-----------|-------| | Max iterations | 5 | | Success gate | 80% pass rate | | Circuit breaker 1 | Score drops >20% on any dimension | | Circuit breaker 2 | 3 consecutive iterations with no improvement | | Rollback mechanism | Git commit after each iteration | ## Iteration Loop ### Step 1: Baseline Evaluation Run evaluation on all dimensions: ```bash python -m tests.evaluation.cli --dimension all --output baseline.json ``` Record the baseline pass rate for each dimension in a TodoWrite checklist. ### Step 2: Identify Worst Dimension From baselin
- When to Use
- Pre-flight Checks
- Guardrails (USER-DEFINED)
- Iteration Loop
- Step 1: Baseline Evaluation
- Step 2: Identify Worst Dimension
- Step 3: Analyze Failures
- Step 4: Edit Prompt
- Step 5: Re-evaluate
- Step 6: Check Circuit Breakers
- Step 7: Commit if Improved
- Step 8: Repeat or Stop
- Output Format
- Quick Reference
curl http://localhost:8004/v1/models ls tests/golden/ git checkout -b eval/prompt-iteration-$(date +%Y%m%d-%H%M) python -m tests.evaluation.cli --dimension all --output baseline.json python -m tests.evaluation.cli --dimension <failing_dimension> --verbose python -m tests.evaluation.cli --dimension <dimension> --output iteration-N.json git add src/backend/app/prompts/ git commit -m "prompt: improve <dimension> pass rate N% -> M% Tested with: python -m tests.evaluation.cli --dimension <dimension> Full evaluation
What does the prompt-iteration skill do?
Use when iteratively improving agent prompts through automated LLM-as-Judge evaluation. Runs eval→fix→commit loop with circuit breakers.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill prompt-iteration --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.