# Keryn Skills

Describes Keryn 0.5.2 and keryn-rules 0.8.0.

> The 8 agent skills in @keryn/skills 1.2.1: what each one does and what it needs, generated from the skill files.

Source: https://keryn.dev/docs/reference/skills/

> **Note: Package version**
> `@keryn/skills` is versioned separately from the CLI. The version on npm is 1.2.1. See [Download](https://keryn.dev/download/#skills).

`@keryn/skills` 1.2.1 installs 8 skills into a project for Claude Code (`/<name>`) and Codex (`$<name>`). Each runs the installed Keryn CLI or MCP server and adds no scanner logic. How to install them and what they change: the [Keryn Skills guide](https://keryn.dev/docs/guides/keryn-skills/). The skill sections on this page are generated from each skill's `SKILL.md`.

## Quickstart

This runs with Node.js 20 or newer and nothing else. It previews the files, writes them into a new directory and lists what was written. `/path/to/` stands for the directory you run it in.

```sh
mkdir demo-project
```

```sh
npx --yes @keryn/skills --target demo-project --host both --dry-run
```

```text
Dry run: /path/to/demo-project
would create .agents/skills/keryn-review/SKILL.md
would create .agents/skills/keryn-diff/SKILL.md
would create .agents/skills/keryn-explain/SKILL.md
would create .agents/skills/keryn-fix/SKILL.md
would create .agents/skills/keryn-verify/SKILL.md
would create .agents/skills/keryn-mcp-audit/SKILL.md
would create .agents/skills/keryn-cloud/SKILL.md
would create .agents/skills/keryn-attribution/SKILL.md
would create .claude/skills/keryn-review/SKILL.md
would create .claude/skills/keryn-diff/SKILL.md
would create .claude/skills/keryn-explain/SKILL.md
would create .claude/skills/keryn-fix/SKILL.md
would create .claude/skills/keryn-verify/SKILL.md
would create .claude/skills/keryn-mcp-audit/SKILL.md
would create .claude/skills/keryn-cloud/SKILL.md
would create .claude/skills/keryn-attribution/SKILL.md
No agent settings, permission rules, global skills, or CLI installation were changed.
```

```sh
npx --yes @keryn/skills --target demo-project --host both
```

```text
Project skills: /path/to/demo-project
created .agents/skills/keryn-review/SKILL.md
created .agents/skills/keryn-diff/SKILL.md
created .agents/skills/keryn-explain/SKILL.md
created .agents/skills/keryn-fix/SKILL.md
created .agents/skills/keryn-verify/SKILL.md
created .agents/skills/keryn-mcp-audit/SKILL.md
created .agents/skills/keryn-cloud/SKILL.md
created .agents/skills/keryn-attribution/SKILL.md
created .claude/skills/keryn-review/SKILL.md
created .claude/skills/keryn-diff/SKILL.md
created .claude/skills/keryn-explain/SKILL.md
created .claude/skills/keryn-fix/SKILL.md
created .claude/skills/keryn-verify/SKILL.md
created .claude/skills/keryn-mcp-audit/SKILL.md
created .claude/skills/keryn-cloud/SKILL.md
created .claude/skills/keryn-attribution/SKILL.md
No agent settings, permission rules, global skills, or CLI installation were changed.
```

```sh
ls demo-project/.claude/skills
```

```text
keryn-attribution
keryn-cloud
keryn-diff
keryn-explain
keryn-fix
keryn-mcp-audit
keryn-review
keryn-verify
```

The skills are now in the project. To do anything they need the Keryn CLI, and `keryn-cloud` needs the Keryn MCP server (see [Limits](#limits)).

## keryn-review

Review a repository with the installed Keryn CLI and explain its findings, AI inventory, policy result, and measured coverage. Use for a requested Keryn security review; code changes require a fix task. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Reads the project; keryn-fix is the one that changes code.

## keryn-diff

Compare Keryn security findings and AI components against an explicit Git or saved-result baseline for a change or CI review. Use when the user asks what security-relevant evidence changed. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Reads the project; keryn-fix is the one that changes code.

## keryn-explain

Explain a specific Keryn rule or saved finding using its source location, trace, proof level, remediation, and limitations. Use when the user needs to understand why a finding matters or how to assess it. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Reads the project; keryn-fix is the one that changes code.

## keryn-fix

Fix a specified Keryn finding in the user's code and check the result with targeted tests and a comparable rescan. Use when a code change is requested, preserving concurrent work and the existing gate policy. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Reads the project; keryn-fix is the one that changes code.

## keryn-verify

Verify a claimed Keryn security fix by comparing before/after evidence and running relevant checks, or verify an evidence bundle's integrity. Use for an explicit verification request; do not equate a clean rescan with exploit reproduction. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Reads the project; keryn-fix is the one that changes code.

## keryn-mcp-audit

Audit the MCP servers a project or this machine would start, judge a server's tools for risk and poisoning, pin them and detect later changes. Use when the user asks whether an MCP server is safe to add or whether one changed. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login). Read-only unless the user asks to pin or wrap.

## keryn-cloud

Run and read Keryn Cloud scans of public addresses through the Keryn MCP server, within the organization's credit. Use when the user asks for a hosted trust, AI-surface or MCP scan, or to check Keryn Cloud credit and results. Needs the Keryn MCP server (npm install -g @keryn/mcp) with KERYN_API_KEY set or a paired device (keryn-mcp pair).

## keryn-attribution

Find where AI coding tools are used in a repository and which commits declare them, with a confidence on every answer. Use when the user asks which AI tools a project uses, how much of its history declares AI involvement, or whether a commit or a file shows AI assistance. Needs the keryn CLI (download from https://keryn.app/download after signing in, then keryn login) and, for the commit signals, a git work tree. Reads file paths and commit metadata; sends nothing anywhere.

## Limits

- The package contains skill files and an installer. It contains no scanner. Every skill needs the Keryn CLI, which is not on npm, and `keryn-cloud` needs the Keryn MCP server with a credential.
- The skills are for Claude Code and Codex. The installer writes no files for other agents.
- It installs into one existing project directory. There is no global install, and both `--target` and `--host` are required.
- It never overwrites. A skill file that differs from the packaged one is refused and nothing is written. There is no force option. An identical file is reported as unchanged.
- It does not edit `AGENTS.md`, `CLAUDE.md`, settings, permission rules or hooks.
- The skills have not been run end to end in a real Claude Code or Codex conversation.
