# Skillhawk > Free, open-source review, security and test checks for AI agent skills. The vault is open source and runs on your own server; hosted Pro is planned. Open source (MIT) command-line tools: grade skills A to F, security-check them and the config around them, measure whether they help, compile them for Claude Code, Codex, Cursor, Copilot and Gemini. ## Key pages - [Home](https://skillhawk.ai/): What Skillhawk does, step by step - [Tools](https://skillhawk.ai/tools/): Every command, with real example output - [Skill library](https://skillhawk.ai/skills/): 277 public skills with grade, findings and security status - [Proof](https://skillhawk.ai/proof/): Scan of 511 skills, lift evals, reviewer accuracy, library grades, with method - [Learn](https://skillhawk.ai/learn/): Sourced answers about skills and instruction files ## Answers - [Why isn't my Claude skill triggering?](https://skillhawk.ai/learn/why-isnt-my-claude-skill-triggering/): The usual causes of a Claude Code skill that never fires, in the order to check them, with a worked example and the commands that find them. - [AGENTS.md vs CLAUDE.md vs Cursor rules: keeping them in sync](https://skillhawk.ai/learn/agents-md-vs-claude-md-vs-cursor-rules/): Which agent reads which instruction file, why copies drift apart, and a low-effort way to keep one source of truth. - [Are agent skills safe? What to check before installing one](https://skillhawk.ai/learn/are-agent-skills-safe/): A skill is text plus optional scripts that your agent runs with your permissions. What to read, what to scan for, and what our scan of 511 popular skills found. - [How to write a SKILL.md description that triggers](https://skillhawk.ai/learn/how-to-write-a-skill-md-description-that-triggers/): The description is the only part of a skill the agent reads before deciding to load it. A formula, a before-and-after, and a check you can run. - [Why does my agent forget rules after compaction?](https://skillhawk.ai/learn/why-does-my-agent-forget-rules-after-compaction/): Long sessions get compacted, and users report rules from CLAUDE.md being ignored afterwards. What is known, what helps, and what Skillhawk does not do. ## Commands - npx skillhawk review