A general-purpose output auditing skill for Claude, Codex, Gemini CLI, and other agents supporting the Agent Skills standard.
Before you deliver a complex document, analysis report, grant section, protocol, roadmap, strategy, or any substantial artifact — harden audits it against domain-specific best practices, classifies issues by severity, fixes them, and re-audits until only cosmetic issues remain.
Existing quality skills are code-focused (verification-before-completion) or single-concern (clarity-gate). harden is the only skill combining web-based research, dual-pass auditing (ideas + structure), severity-classified fix execution, and iterative convergence — and it works on any deliverable type, not just code.
- Research — Searches the web for domain-specific best practices before auditing
- Dual-pass audit — Checks both ideas (are they sound?) and structure (do paths/references/formatting work?)
- Severity classification — 🔴 Critical / 🟡 Significant / 🟢 Minor with specific locations
- Fix — Implements all critical and significant fixes (not just reports)
- Converge — Re-audits until only minor issues remain (max 3 passes)
npx skills add mickeylorenzini/harden
skillkit install mickeylorenzini/harden
Copy SKILL.md to your agent's skills directory. Common paths:
- Claude Code:
~/.claude/skills/harden/SKILL.md - Codex CLI:
~/.codex/skills/harden/SKILL.mdor~/.agents/skills/harden/SKILL.md(check Codex docs for your version) - Gemini CLI:
~/.gemini/skills/harden/SKILL.md
- claude.ai / Claude Desktop: Download this repo as ZIP → Settings → Capabilities → Skills → Upload
- Other chat apps: Upload SKILL.md to your conversation or install via custom instructions in settings
Works with any LLM that can follow structured instructions.
Activates automatically on substantial deliverables.
In CLI tools, trigger manually with /harden.
In chat interfaces, say "harden this" or "audit this before I finalize."
Built by Mickey Lorenzini — Staff scientist, Salk Institute for Biological Studies.