Agent Skills
Task-specific Agent Skills give coding agents (and humans) focused guidance for safety-sensitive or convention-heavy work. They live in .skills/ — one directory per skill, each with a SKILL.md — and are exposed to coding agents through .agents/skills/, managed by skilld.
Shipped skills
| Skill | Use it when… |
|---|---|
agent-zero-architecture | Adding features, moving code, or changing dependencies across packages and adapters. |
agent-zero-cli | Adding or changing CLI commands, flags, prompts, output, or exit codes. |
agent-zero-safety | Touching the runner, autonomous state transitions, repository writes, command execution, secrets, or observe/fix policy. |
orpc-server | Changing dashboard server routes, oRPC contracts, handlers, middleware, or typed clients. |
tsdown | Building or changing packages with the shared tsdown configuration. |
turborepo-monorepo | Changing workspace scripts, turbo.json tasks, caching, or CI orchestration. |
vitehub | Working with ViteHub server primitives, Runtime Helpers, or the Nuxt/Nitro integration. |
Commands
aube run skills:list # show skilld-managed project skills
aube run skills:install # restore skilld links
aube run check:repo # verify skills and community files
Conventions
aube run check:repo (backed by scripts/check-repository.mjs) enforces the skill conventions in CI:
- every
.skills/<name>/directory has aSKILL.md; - the frontmatter
namematches the directory name anddescriptionis present; SKILL.mdstays under 500 lines;- content passes a prompt-injection and unsafe-pattern scan;
- each skill is linked into
.agents/skills/<name>.
AGENTS.md and .github/copilot-instructions.md provide concise entry points for coding agents without replacing the human contributor guide. The pull-request rule in AGENTS.md applies to skills too: update documentation, examples, and skills in the same change that alters commands, boundaries, or contributor workflows.
Formatting and linting
Oxfmt owns formatting and Oxlint owns linting. ESLint and Prettier are deliberately absent — do not add them.
Architecture
Agent Zero is organized as a dependency-directed monorepo. The core decides what should happen; adapters decide how external systems communicate with it; the runner controls what is allowed to happen to a checkout.