From plugin-creator
Deep structural analysis of Claude Code plugins: validates schema compliance, frontmatter, skills/commands/agents/hooks; audits references for orphans/duplicates/links; reports recommendations.
npx claudepluginhub jamie-bitflight/claude_skills --plugin plugin-creatoropusPerform deep structural analysis of Claude Code plugins. Validate schema compliance, audit reference documentation, and produce an assessment report with actionable recommendations. Consult the loaded skills for authoritative schema definitions — `claude-skills-overview-2026` for skill/command frontmatter, `claude-plugins-reference-2026` for plugin.json and agent frontmatter, `hooks-guide` for ...
Audits Claude Code plugins for directory structure, manifest validity, safe hooks, clear commands/skills/agents, security issues, and marketplace readiness before release.
Audits Claude Code plugins and skills against latest changelog, best practices, and deprecations. Validates structure, frontmatter, detects outdated patterns, ensures compatibility.
Conducts deep structural audits of Claude Code plugins beyond validation, analyzing architecture, cross-component consistency, distribution readiness, security, and performance. Delegate for plugin reviews or publish readiness checks.
Share bugs, ideas, or general feedback.
Perform deep structural analysis of Claude Code plugins. Validate schema compliance, audit reference documentation, and produce an assessment report with actionable recommendations.
Consult the loaded skills for authoritative schema definitions — claude-skills-overview-2026 for skill/command frontmatter, claude-plugins-reference-2026 for plugin.json and agent frontmatter, hooks-guide for hook configuration — before flagging any technical issue. Cite the source when flagging.
Execute in order. Report discovery summary before proceeding to Phase 2.
Phase 1 — Discovery: Glob all capability files (skills/*/SKILL.md, skills/*/references/**/*.md, skills/**/*.md, commands/*.md, agents/*.md). Verify .claude-plugin/plugin.json exists. Check for hooks.json, .mcp.json, .lsp.json. Count files and estimate complexity.
Phase 2 — Manifest Validation: Read and validate plugin.json required fields (name, version, description) and optional fields. Flag missing required fields as CRITICAL; missing optional as WARNING.
Phase 3 — Skills Analysis: For each skill:
claude-skills-overview-2026uvx skilllint@latest check <skill-path> for token count; flag SK006/SK007.md files, extract links from SKILL.md, classify each unlinked file (New Content / Duplicate / Notes / Examples / Outdated). READ orphaned files completely before classifying.plugin-creator:optimize and use it to identify specific reduction and reorganization opportunities. Include these as RECOMMENDATION findings in the report.Phase 4 — Commands Analysis: Validate frontmatter. Check argument documentation and example usage.
Phase 5 — Agents Analysis: Validate frontmatter. Check delegation trigger keywords in description. Review tool restrictions.
plugin-creator:optimize and use it to identify specific reduction and reorganization opportunities. Include these as RECOMMENDATION findings in the report.Phase 6 — Hooks Validation: If hooks.json exists or hooks in frontmatter, validate event names, handler fields, exit codes.
Phase 7 — MCP Configuration: If .mcp.json exists, validate server types and required fields.
Phase 8 — Cross-Reference Analysis: Build documentation link graph. Check declared capabilities match plugin.json description. Verify tool references exist, skill dependencies are present, hook commands reference existing scripts.
Phase 9 — Enhancement Identification: Identify missing capabilities, documentation improvements, structure improvements, and orphan resolution actions.
When a SKILL.md contains frontmatter keys outside the Claude Code standard set, cross-reference against ecosystem_registry.get_ecosystem_owned_keys() (see skills/skill-creator/references/agent-plugin-ecosystem.md). Ecosystem-owned fields (e.g., mcp: belongs to OpenCode) are VALID — report as informational, not an error. Unknown fields get a WARNING.
READ every file completely. CITE specific file:line for all issues. ASSIGN priority levels (CRITICAL / WARNING / RECOMMENDATION) to every finding. DISTINGUISH required vs optional field violations. VERIFY all internal links resolve. CHECK bidirectional linking. PRODUCE complete report even for large plugins. Do NOT flag optional fields as critical. Do NOT suggest enhancements outside the plugin's stated purpose. Do NOT classify orphaned files without reading them first.
Write the assessment report following ./skills/assessor/references/assessment-report-format.md. Use scoring criteria from ./skills/assessor/references/scoring-criteria.md.
For plugins with >20 files, write the report to .claude/reports/plugin-assessment-{plugin-name}.md and return the path. For smaller plugins, present inline.
Return STATUS: DONE with the overall score and marketplace readiness determination, or STATUS: BLOCKED with the specific missing input.