Skip to content
cccskills
FreePublished on GitHub

dev:validate

Use when the user wants to validate the plugin, check integrity, verify cross-references, or before a release. Also use when the user says 'validate', 'check plugin', or 'verify'.

The English summary is being prepared. Showing the original description.

See how to install

Files included (1)

  • SKILL.md3.6 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

/dev:validate

Check plugin integrity: skill/agent frontmatter, cross-references, Python tool syntax, version sync, hook script executability.

Checks

1. Version Sync

PKG_V=$(python3 -c "import json; print(json.load(open('package.json'))['version'])")
PLUGIN_V=$(python3 -c "import json; print(json.load(open('.claude-plugin/plugin.json'))['version'])")
if [ "$PKG_V" = "$PLUGIN_V" ]; then
    echo "OK: versions match ($PKG_V)"
else
    echo "FAIL: package.json=$PKG_V, plugin.json=$PLUGIN_V"
fi

2. Skill Frontmatter

For each skills/*/SKILL.md:

  • Must have name: in frontmatter
  • Must have description: in frontmatter
  • Must have allowed-tools: in frontmatter
for f in skills/*/SKILL.md; do
    NAME=$(grep -m1 "^name:" "$f" | cut -d: -f2- | xargs)
    DESC=$(grep -m1 "^description:" "$f")
    TOOLS=$(grep -m1 "^allowed-tools:" "$f")
    if [ -z "$NAME" ] || [ -z "$DESC" ] || [ -z "$TOOLS" ]; then
        echo "FAIL: $f missing frontmatter fields"
    else
        echo "OK: $f ($NAME)"
    fi
done

3. Agent Frontmatter

For each agents/*.md:

  • Must have name: in frontmatter
  • Must have description: in frontmatter
  • Must have tools: in frontmatter
  • Must have color: in frontmatter
for f in agents/*.md; do
    NAME=$(grep -m1 "^name:" "$f" | cut -d: -f2- | xargs)
    COLOR=$(grep -m1 "^color:" "$f" | cut -d: -f2- | xargs)
    if [ -z "$NAME" ] || [ -z "$COLOR" ]; then
        echo "FAIL: $f missing name or color"
    else
        echo "OK: $f ($NAME, $COLOR)"
    fi
done

4. Agent Cross-References

Check that every subagent_type: referenced in skills exists as an agent file:

for AGENT in $(grep -roh 'subagent_type: "[^"]*"' skills/ | sed 's/subagent_type: "//;s/"//' | sort -u); do
    if [ ! -f "agents/$AGENT.md" ]; then
        echo "FAIL: subagent_type '$AGENT' referenced in skills but agents/$AGENT.md not found"
    else
        echo "OK: $AGENT agent exists"
    fi
done

5. Python Tool Syntax

ERRORS=0
for f in tools/*.py; do
    python3 -c "import ast; ast.parse(open('$f').read())" 2>&1
    if [ $? -ne 0 ]; then
        echo "FAIL: $f has syntax errors"
        ERRORS=$((ERRORS+1))
    else
        echo "OK: $f"
    fi
done
echo "Python tools: $ERRORS errors"

6. Hook Script

if [ -f "hooks/session-start.sh" ]; then
    if [ -x "hooks/session-start.sh" ]; then
        echo "OK: hooks/session-start.sh is executable"
    else
        echo "FAIL: hooks/session-start.sh not executable"
    fi
    if [ -f "hooks/hooks.json" ]; then
        python3 -c "import json; json.load(open('hooks/hooks.json'))" 2>&1
        if [ $? -eq 0 ]; then echo "OK: hooks.json valid JSON"; else echo "FAIL: hooks.json invalid"; fi
    fi
else
    echo "WARN: no hooks/session-start.sh"
fi

7. CLAUDE.md Accuracy

Check that tool count and agent count in CLAUDE.md match reality:

TOOL_COUNT=$(ls tools/*.py 2>/dev/null | wc -l)
AGENT_COUNT=$(ls agents/*.md 2>/dev/null | wc -l)
echo "Tools: $TOOL_COUNT Python files"
echo "Agents: $AGENT_COUNT agent definitions"

Report

Print a summary:

Plugin Validation:
  Versions: {OK/FAIL}
  Skills: {N} checked, {N} passed
  Agents: {N} checked, {N} passed
  Cross-refs: {N} checked, {N} passed
  Python tools: {N} checked, {N} syntax errors
  Hooks: {OK/FAIL}

Result: {PASS/FAIL}

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

Use when the user wants to smoke-test the evolve pipeline, test tools, or verify the plugin works end-to-end. Also use when the user says 'dry run', 'smoke test', or 'test pipeline'.

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

Use when the user wants to release a new version, publish to npm, create a GitHub release, bump version, or tag a release. Also use when the user says 'release', 'publish', 'bump version', 'tag', or 'npm publish'.

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

Use when the user wants to verify that the evolved agent's score is stable and reliable. Runs evaluation multiple times and reports mean ± std.

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

Use when the user is done evolving and wants to finalize, clean up, tag the result, or push the optimized agent.

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

Use when the user wants to run the optimization loop, improve agent performance, evolve the agent, or iterate on quality. Requires .evolver.json to exist (run harness:setup first).

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

Use when the user wants to check dataset quality, diagnose eval issues, or before running evolve. Checks size, difficulty distribution, dead examples, coverage, and splits. Auto-corrects issues found.

The English summary is being prepared. Showing the original description.

raphaelchristi/harness-evolver56Updated Apr 18, 2026

See all skills by raphaelchristi

Report a problem with this skill