Agent Skills for building on Orca Agent Engine, the open-source runtime for managed AI agents. The engine's registry API is compatible with Anthropic's Managed Agents API.
| Skill | Description |
|---|---|
orca-api |
Teaches an AI coding agent how to work with the Agent Engine registry API, the ork CLI, and the @runorca/orca-sdk TypeScript SDK: the resource model (agents, sessions, environments, files, skills, vaults, memory stores, triggers), authentication, events and streaming, pagination, and the error model. |
In Claude Code, add this repository as a plugin marketplace and install the skill:
/plugin marketplace add orca-ae/orca-skills
/plugin install orca-api@orca-skills
For a local checkout, use the path form instead:
/plugin marketplace add /path/to/orca-skills
Then describe what you want to build, for example "set up an Orca agent that triages new GitHub issues". The skill loads when a prompt or project mentions Orca, Agent Engine, ork, @runorca/orca-sdk, or the ORCA_* variables. Other agents that read the Agent Skills format can load skills/orca-api directly.
What the skill's code needs at run time:
- An Agent Engine deployment.
ork local startruns one on your machine with Docker. Export your model provider's key before you start it —ANTHROPIC_API_KEYfor the Claude models the skill defaults to,OPENAI_API_KEYfor OpenAI models — or the first session fails to authenticate. The engine's quick start covers the other ways to run it, or use a hosted deployment. - The
orkCLI for setup scripts and CI:brew install orca-ae/tap/ork. - The TypeScript SDK for application code:
npm install @runorca/orca-sdk.
- Engine documentation, including which engine,
ork, and SDK versions work together - Cookbooks: complete programs that run end to end
- runorca.ai
- Questions and ideas: Discussions in this repository.
- The skill says something the engine doesn't do: open an issue here.
- Bugs in the engine,
ork, or the TypeScript SDK: open an issue on the engine repository. - Security problems: report them privately, as SECURITY.md describes, never in a public issue.
CONTRIBUTING.md covers the workflow: DCO sign-off, no CLA, and how to disclose AI assistance under the AI policy. The verification gates run against published artifacts only:
npm ci --prefix scripts # the pinned SDK, TypeScript and Node types
brew install orca-ae/tap/ork # or set ORK_BIN to an ork binary
scripts/check-all.sh # every gate; run it before each commit| Gate | What it checks |
|---|---|
scripts/check-hygiene.sh |
names and links across the whole tree: renamed tools, private repositories, personal paths |
scripts/check-skill.sh |
skill claims that don't match Orca: upstream-only surfaces, absent endpoints, wrong CLI shapes |
scripts/check-ts-snippets.sh |
typechecks every TypeScript example against the published @runorca/orca-sdk |
scripts/check-cli-commands.sh |
every documented ork command and flag against ork --help, and the stream output shape |
scripts/check-xrefs.sh |
cross-references, frontmatter, the manifest, and the engine and SDK paths the skill cites |
scripts/check-license-headers.mjs |
the license header on every script and workflow (--fix adds it) |
The gates need Node.js 20 or newer with npm, ork, python3 (for a local SSE fixture server), and jq. To check the engine paths the skill cites, point ORCA_ENGINE_DIR at a clone of orca-ae/orca-agent-engine; ORCA_REQUIRE_LIVE_SOURCES=1 turns a missing clone or SDK install into a failure.
Apache License 2.0 — see LICENSE and NOTICE.
The orca-api skill is a derivative of the claude-api skill from anthropics/skills, Copyright 2026 Anthropic, PBC, used under the Apache License 2.0 (retained at skills/orca-api/LICENSE.txt). The files have been modified to target Orca Agent Engine, the ork CLI, and the @runorca/orca-sdk TypeScript SDK.