Agents and skills
Cabin ships AI-agent affordances so coding agents (Cursor, Gemini, Claude, and others) implement against the same contracts humans read.
Affordances
| Artifact | Role |
|---|---|
| AGENTS.md | Authoritative hard rules for any agent in this repo |
| llms.txt | Concise LLM-oriented doc map and reading order |
| skills/ | Modular Agent Skills (SKILL.md per concern) |
| .cursor/rules/cabin.mdc | Cursor project rules mirroring hard constraints |
Skills layout
Follows the open Agent Skills pattern (and is aligned in spirit with android/skills):
skills/
cabin-overview/SKILL.md
cabin-tokens/SKILL.md
cabin-compliance/SKILL.md
cabin-compose/SKILL.md
cabin-views/SKILL.md
cabin-soong/SKILL.md
cabin-component-spec/SKILL.md
Each SKILL.md has YAML frontmatter (name, description with “use when…”)
plus a focused body. Agents should auto-select by description; humans can
point agents at a skill path explicitly.
How humans use them
- Point the agent at AGENTS.md (many tools load this automatically).
- For a scoped task, attach or invoke the relevant skill under
skills/. - Prefer docs linked from llms.txt over inventing architecture.
- Review agent PRs against pillars and hard rules — agents do not replace review (contributing).
Relationship to android/skills
android/skills provides AI-optimized modular Android instructions from Google. Cabin:
- Grounds engineering principles in those modern Android patterns (principles)
- Adds Cabin-specific skills for automotive compliance, dual-stack parity, tokens, and Soong/build-tree adoption — concerns android/skills does not encode for this design system
Use both: android/skills for general Android/Compose craft; Cabin skills for product and platform constraints.