| name | rules-agents |
| description | Use when installing rules_agents into a Bazel repo, verifying the setup, discovering skills from registries or custom repos, adding local skills, and configuring agent profiles and runners. |
rules_agents
Use this skill for repo-scoped agent setup in Bazel. Keep the flow narrow:
- supported runners:
codex, claude_code, pi
- supported host platforms: Linux and macOS; Windows is not supported yet
- repo-local install roots:
.agents/skills, .claude/skills, .pi/skills
- public API:
agent_skill, agent_profile, agent_runner, skill_deps
Discoverable Parts
- For the fastest install-and-verify path, read
references/quickstart.md.
- For ongoing usage, read
references/usage.md.
Working Rules
- Prefer the quickstart path first unless the user already has a profile and just needs a specific usage flow.
- Do not assume
rules_agents is published to the Bazel Central Registry. Use the quickstart source override path until a released BCR module is available.
- Use registry discovery only to browse available skills. A discovered skill becomes usable only after adding the printed
skill_deps.remote(...) snippet and use_repo(...).
- Use
agent_skill(...) for repo-local skills and skill_deps.remote(...) for remote archives.
- Use
agent_profile(...) to collect skills and required credential env vars.
- Use
agent_runner(...) to pick codex, claude_code, or pi and expose :name, :name_doctor, :name_setup, :name_run, and :name_manifest.
Verification Defaults
bazel run //:dev_doctor verifies binary resolution, required env vars, and packaged skills without launching.
bazel build //:dev_manifest verifies manifest generation.
bazel run //:dev -- --help is a safe launch-path check once doctor passes.