Skip to main content

Agents and skills

Cabin ships AI-agent affordances so coding agents (Cursor, Gemini, Claude, and others) implement against the same contracts humans read.

Affordances

ArtifactRole
AGENTS.mdAuthoritative hard rules for any agent in this repo
llms.txtConcise LLM-oriented doc map and reading order
skills/Modular Agent Skills (SKILL.md per concern)
.cursor/rules/cabin.mdcCursor 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

  1. Point the agent at AGENTS.md (many tools load this automatically).
  2. For a scoped task, attach or invoke the relevant skill under skills/.
  3. Prefer docs linked from llms.txt over inventing architecture.
  4. 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.