# Keryn Skills

> Keryn Skills (@keryn/skills): eight agent skills that connect Claude Code and Codex to the Keryn CLI and the Keryn MCP server.

Source: https://keryn.dev/docs/guides/keryn-skills/

If you work with a coding agent, eight project skills (Keryn Skills) connect it to the installed Keryn CLI and MCP server: `keryn-review` (a scope), `keryn-diff` (a change against an explicit base), `keryn-explain` (a finding's evidence and uncertainty), `keryn-fix` (a scoped correction, relevant tests and a comparable rescan), `keryn-verify` (a claim about a correction), `keryn-mcp-audit` (the MCP servers a project or machine uses), `keryn-cloud` (scans in Keryn Cloud through the MCP server) and `keryn-attribution` (where AI coding tools are used and which commits declare them). In Claude Code they are `/keryn-review` and so on; in Codex they are `$keryn-review`. They scan and explain with the CLI's own outputs and add no scanner logic; the agent contributes code reading, suggested changes and your project's tests.

> **Note: Install from npm**
> The skills ship as the package `@keryn/skills`, available from npm: `npx @keryn/skills --target <project> --host both --dry-run` previews the files, and without `--dry-run` it writes `.claude/skills/keryn-*/SKILL.md` and `.agents/skills/keryn-*/SKILL.md` into that project only. It overwrites nothing, edits no `AGENTS.md`, `CLAUDE.md`, settings or hooks, and reports a conflict instead of replacing a file you changed. Installing the skills does not install the scanner. They have not been run end to end in a real Claude or Codex conversation.

The scan stays on your machine. What the agent host passes to its own model provider follows that host's settings, and Keryn does not control it. A model's explanation does not raise a finding's proof level, and a finding that disappears is not on its own proof of a fix.

See also the [Keryn MCP](https://keryn.dev/docs/reference/mcp-tools/) server the skills call, and the [Keryn SDK](https://keryn.dev/docs/reference/sdk/).
