Agent Skills Reference
| Skill | Trigger | Done when |
|---|---|---|
orcats-setup | Install or verify Orca. | orcats --version succeeds and at least one backend is ready or unverified. |
orcats-author | Create a saved workflow or loop. | The artifact is saved, gated, and typechecked when possible. |
orcats-flow | Run or heal a saved artifact. | The run completes or escalates with classification, evidence, and safe next steps. |
Installation command
Section titled “Installation command”orcats skills [--list] [--skill <name>|--all] [--agent <name>] [--global] [--yes]
delegates to npx skills add ASRagab/orca-ts. It requires Node.js/npm, keeps
the installer interactive when no selection is provided, and does not own
agent-directory placement, updates, or removal. See the
installation guide for examples and direct
npx skills fallback commands.
orcats-setup installs or locates the standalone orcats binary, asks which backend to enable, runs the bundled doctor, and classifies missing, unauthenticated, or misconfigured CLIs.
Author
Section titled “Author”orcats-author reads the target repo, detects real test and lint commands, interviews for workflow shape, fills a checked template, and saves either .orca/workflows/<name>.ts or .orca/loops/<name>.ts.
Mutating artifacts must include verification gates and the shared baseline policy. The default is repair; strict and accept-dirty are explicit overrides via --baseline=<policy> or ORCA_BASELINE_POLICY.
orcats-flow runs saved artifacts, watches monitoring JSON, loop state, persistent plans, and git progress, then diagnoses backend failures, baseline repair progress, gate failures, non-convergence, stalls, crashes, and served-child failures. It does not retry with accept-dirty unless the operator explicitly asks.