Skills

Skills are slash commands — reusable, multi-step workflows Claude executes on
demand. They are plain markdown files that Claude reads and follows as
instructions.

File Format

Skills live in _protocol_/.claude/skills/<name>/SKILL.md:

_protocol_/.claude/skills/
  release/
    SKILL.md
  deploy/
    SKILL.md

The skill name is the directory name. Invoke it with /release, /deploy, etc.

A SKILL.md is a markdown document describing what Claude should do step by
step. No special syntax — just clear instructions.

Included Skills

The protocol ships with two skills:

• /release — full release workflow: version bump, changelog, build
  verification, tag, push, GitHub release, MDPlanner progress note.
• /import-project — onboard a codebase into MDPlanner and spawn its brain
  in one automated flow. Accepts NAME=MyApp PROJECT=/path/to/code as
  arguments. Run this from the cerveau.dev root session to bootstrap a new
  project.

Read the SKILL.md files in _protocol_/.claude/skills/ for the full
procedures and use them as templates when writing your own.

Writing Skills

Skills work best for workflows you run repeatedly that have a fixed sequence
of steps. Examples:

Create a skill for deploying to staging.
Steps: build, run smoke tests, push image to registry, update the
deployment, verify health endpoint returns 200.
Save it to _protocol_/.claude/skills/deploy-staging/SKILL.md

Create a skill for cutting a hotfix.
Steps: checkout main, create hotfix branch, confirm the fix is committed,
bump patch version, changelog entry, tag, push, PR.
Save it to _protocol_/.claude/skills/hotfix/SKILL.md

Keep skills focused. One workflow per skill. If a skill branches based on
conditions, split it into two skills.

Skills vs. Agents

	Skills	Agents
Invocation	/skill-name by user	Called by Claude when needed
Purpose	Multi-step workflow	Specialized sub-instance
Format	Plain markdown steps	YAML frontmatter + system prompt
Tool access	Inherits all tools	Restricted to declared tools