peer-ai-skills

Peer AI's skills, in the Agent Skills format that Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI all read. Each skill does a job and proves it: it builds an inventory of what it's checking, checks every rule against it, and hands its output to Peer AI, which checks it in turn. RFC 0004 sets out the design.

You don't install this package yourself. peer-ai render writes the skills where each of your AI tools reads them.

What's here

| Path | What it holds | |------|---------------| | skills/<id>/ | Each skill's source: SKILL.md, its own references, a document skill's templates in assets/, and Codex's agents/openai.yaml | | shared/ | The severity scale and report format every review skill gets | | src/build.ts | Builds a skill: adds its rules from peer-ai-standards and the shared references, and sets the name it's written under | | src/validate.ts | Checks a built skill against the specification, Anthropic's and OpenAI's guidance, and RFC 0004 | | src/document.ts | Compares a document with its skill's template: the check behind the check_document tool | | AUTHORING.md | How to write a skill, and which rules the build enforces |

The three kinds

| Kind | Output | Checked by | |------|--------|------------| | Review | A review report (RFC 0002) | The record_review tool, or peer-ai check-report | | Document | A document from a template | The check_document tool, or peer-ai check-document | | Work | Work items created and moved | The work item's gates |

Every skill has at least three evals before it ships. See AUTHORING.md.