Review Skill

Review Skill Workflow

You are helping an SME review an Agent Skill before publishing. This is a multi-step process: determine environment, verify prerequisites, run structural validation, review content, optionally run LLM scoring, and interpret results. Follow every step in order.

Step 0: Determine Environment

Check for saved configuration:

cat ~/.config/skill-validator/review-state.yaml 2>/dev/null

If the state file exists with prereqs_passed: true, offer:

Found saved settings — configured for [full/structural-only] reviews.

1. Continue with saved settings — skip to Step 2
2. Re-run prerequisite checks
3. Change environment — switch between full and structural-only

Option 1: read llm_scoring from the file and skip to Step 2. Options 2-3: continue below.

If no state file exists, or the user chose to re-check/change, ask:

LLM scoring evaluates content quality across multiple dimensions.

1. Yes, run LLM scoring — full review with LLM scoring
2. No, skip LLM scoring — structural validation only

Option 1: set LLM_SCORING=true. Option 2: set LLM_SCORING=false. Run Step 1a only, then jump to Step 2.

Step 1: Verify Prerequisites

1a. Check for skill-validator binary

skill-validator --version

If not found, search common locations (/usr/local/bin, /opt/homebrew/bin, ~/go/bin). If found but not on PATH, tell the user. If not found anywhere, follow references/install-skill-validator.md (opens in a new tab).

If --version is not at least v1.5.1, help the user upgrade with brew upgrade skill-validator or go install github.com/agent-ecosystem/skill-validator/cmd/skill-validator@latest.

Do NOT proceed until this succeeds.

1b. Check for claude CLI (LLM scoring only)

If LLM_SCORING=true, verify the Claude CLI is available:

claude --version

If not found, tell the user to install Claude Code:

• macOS: curl -fsSL https://claude.ai/install.sh | bash
• Other platforms: follow the Claude Code quickstart guide (opens in a new tab)

The user must authenticate by running claude interactively before continuing.

Do NOT proceed with LLM scoring until this succeeds.

Save state after prerequisites pass

Persist state so future runs skip this step. Replace <true or false> with the actual LLM_SCORING value:

mkdir -p ~/.config/skill-validator
cat > ~/.config/skill-validator/review-state.yaml << 'EOF'
prereqs_passed: true
llm_scoring: <true or false>
EOF

Step 2: Locate the Skill

Ask the user for the path to the skill they want to review, unless they have already provided it. Verify the path contains a SKILL.md file:

ls <path>/SKILL.md

If SKILL.md does not exist at the given path, tell the user this is not a valid skill directory and ask them to provide the correct path.

Step 3: Run Structural Validation

Run the full check suite:

skill-validator check <path>

Capture the exit code:

Exit 0: proceed. Exit 2: note warnings, proceed. Exit 1: list errors — these are blocking. The user must fix them before the skill can be published. Do NOT proceed to LLM scoring if exit code is 1.

Step 4: Content Review

Read the SKILL.md and any reference files, then evaluate each check below. Report which checks pass and which do not, with specific details on what is missing.

Flag any failing checks as areas the SME should address. These are not blocking but should be resolved before publishing for best results.

Step 5: LLM Scoring and Interpretation

If LLM_SCORING=false, skip to Step 6.

If LLM_SCORING=true, follow the “Run LLM Scoring” and “Interpret LLM Scores” sections of references/llm-scoring.md (opens in a new tab).

Step 6: Present the Review Summary

If LLM_SCORING=true, follow the “Full Review Summary” section of references/llm-scoring.md (opens in a new tab). Include any failing content review checks from Step 4 in the action items.

If LLM_SCORING=false, present structural result, content review result, areas to address, and a self-assessment checklist using the scoring dimensions from assets/report.md (opens in a new tab). Note that LLM scoring was skipped; advise re-running with LLM scoring enabled or self-assessing against the report dimensions.

Example Review Summary Structure

Structure the final summary with these sections in order:

1. Structural validation — pass/fail with errors or warnings
2. SKILL.md scores — overall and per-dimension table
3. Reference scores — per-file table with overall and lowest dimension
4. Novelty assessment — mean novelty vs threshold of 3; list novel_info per file for SME verification
5. Action items — prioritized list of what to fix
6. Recommendation — ready to publish / minor revisions / significant rework