Audit and Optimize Your Skill Ecosystem

Skill Sentinel is a meta-agent that monitors, audits, and evolves an entire skill ecosystem, analyzing every skill across 7 weighted dimensions and generating gap-analysis recommendations for new skills. Code Quality (20% weight) checks cyclomatic complexity per function (threshold 10), function size (threshold 50 lines), file size (threshold 500 lines), docstring coverage, and error-handling patterns (bare/broad except). Security (20%) scans for hardcoded secrets (tokens, passwords, API keys), SQL injection via f-strings in queries, insecure HTTP URLs, tokens leaking into logs, and input validation. Performance (15%) checks for retry-with-backoff on API calls, configured timeouts, HTTP connection reuse, N+1 query patterns, and async/concurrency usage. Governance (15%) rates each skill on a 5-level maturity scale from none through action logging, logging plus rate limiting, full two-step confirmations, to advanced alerting and trend tracking. Documentation (15%) checks SKILL.md frontmatter completeness (name/description/version), bilingual trigger keywords, required/recommended sections, and reference files. Dependencies (15%) verifies a requirements.txt exists with pinned versions and cross-checks imported vs listed dependencies. Cross-Skill analysis (global, unweighted) identifies duplicated modules between skills, shared database patterns, inconsistent governance levels, and extraction opportunities. Beyond the 7 dimensions it runs cost-impact analysis (SKILL.md token size per activation, large unindexed reference files, verbose script output, absence of structured JSON output) and a gap-analysis recommender that compares ecosystem coverage against a 20-category taxonomy and generates ready-to-use SKILL.md templates for suggested new skills. The tool practices its own governance principles: every audit is logged to an action log, score history is tracked over time for trend analysis, and reports are saved to a reports directory. Its CLI (run_audit.py) supports a full ecosystem audit, a single-skill audit (--skill <name>), recommendation-only mode (--recommend), trend comparison against the previous audit (--compare), JSON output for processing, and audit history review. Reports contain an executive score summary, trend deltas, findings by severity (critical/high/medium/low/info), detailed per-skill analysis, new-skill recommendations, and a prioritized action plan.

Use it when auditing skill ecosystem quality, checking skill security across the whole library, verifying overall ecosystem health, finding duplicated skills, or optimizing the skill set as a whole - it operates at the ecosystem level, not as a single-skill activation or security-only reviewer. Do not use it for tasks unrelated to skill-ecosystem auditing, when a simpler specific tool would handle the request, or for general-purpose assistance outside its domain.

Inputs: the skill directory to audit (all skills by default, or --skill <name> for one), plus optional flags for recommendations-only, comparison mode, or JSON output.
Outputs: a structured markdown report with an executive score summary, trend deltas, severity-ranked findings, per-skill detailed analysis, new-skill recommendations with ready templates, and a prioritized action plan.

python run_audit.py --skill instagram
python run_audit.py --recommend
python run_audit.py --compare

Teams managing a large or growing library of agent skills who need ecosystem-wide visibility into code quality, security, performance, governance, documentation, dependency hygiene, duplication, and coverage gaps - not just a single skill's health.