Skip to content
Back to search

claw-orchestrator

Run Claude Code, Codex, Antigravity, Cursor Agent and OpenCode as one runtime — persistent sessions, multi-agent councils, an OpenAI-compatible endpoint, an MCP server, and an ACP agent any editor can drive.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Enderfga/claw-orchestrator with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Enderfga/claw-orchestrator'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/enderfga--claw-orchestrator/badge.svg)](/listings/enderfga--claw-orchestrator)

HTML

<a href="/listings/enderfga--claw-orchestrator"><img src="/api/listings/enderfga--claw-orchestrator/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

npm install -g @enderfga/claw-orchestrator
clawo serve   # dashboard at http://127.0.0.1:18796/dash

The server generates an access token at `~/.openclaw/server-token`; open `http://127.0.0.1:18796/login?token=<token>&redirect=/dash` once to sign in the browser.

import { SessionManager } from '@enderfga/claw-orchestrator';

const manager = new SessionManager();
await manager.startSession({ name: 'fix-tests', engine: 'claude', cwd: '/project' });
const result = await manager.sendMessage('fix-tests', 'Fix the failing tests');

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 253 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <p align="center"><img src="./assets/teaser.webp" alt="Claw Orchestrator: one run through six steps — declare the checks, dispatch to any engine, run, verify, record, verdict — with a failed check sent back to the agent, on a family of coding agents" width="100%"></p>
  2. <h1 align="center">Claw Orchestrator: A Unified Runtime for Coding Agent CLIs</h1>
  3. <p align="center">
  4. 📦 <a href="https://www.npmjs.com/package/@enderfga/claw-orchestrator">npm</a> •
  5. 📖 <a href="./skills/references/getting-started.md">Documentation</a> •
  6. 🎬 <a href="./examples/promo-video/">Demo film</a> •
  7. 📝 <a href="./CHANGELOG.md">Changelog</a> •
  8. 📚 <a href="#citation">Citation</a>
  9. </p>
  10. <p align="center">
  11. <a href="https://www.npmjs.com/package/@enderfga/claw-orchestrator"><img src="https://img.shields.io/npm/v/@enderfga/claw-orchestrator.svg" alt="npm version"></a>
  12. <a href="https://www.npmjs.com/package/@enderfga/claw-orchestrator"><img src="https://img.shields.io/npm/dm/@enderfga/claw-orchestrator.svg" alt="npm downloads"></a>
  13. <a href="https://github.com/Enderfga/claw-orchestrator/actions/workflows/ci.yml"><img src="https://github.com/Enderfga/claw-orchestrator/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
  14. <a href="./LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
  15. </p>
  16. **Claw Orchestrator** runs Claude Code, Codex, Antigravity, Grok Build, OpenCode and any
  17. [Agent Client Protocol](https://agentclientprotocol.com) agent (Kimi Code, dsh) as persistent sessions behind one API.
  18. A run counts as done only when checks the agent did not write pass: you declare the checks, the runtime runs them
  19. and reads the exit codes, and the evidence stays on disk. Runs are checkpointed workflows that survive a crash, and
  20. every turn on every engine is recorded with its model, tokens and cost.
  21. <p align="center">
  22. <img src="./assets/hero.gif" alt="An agent reports that all tests pass; the runtime's own check refutes it. Five engines, three runs each, tallied by verdict." width="800">
  23. </p>
  24. <p align="center"><sub>
  25. One bug, one acceptance contract, five engines, three runs each, first attempt only. The runtime ran the checks itself:
  26. 9 verified, 4 refuted, 2 errored (Grok Build hit its free-tier usage limit). One task, not a benchmark.
  27. <a href="./examples/promo-video/">How the film was made, and how to re-run it</a>.
  28. </sub></p>
  29. <details>
  30. <summary><b>The 40-second film, with sound</b></summary>
  31. https://github.com/user-attachments/assets/5bf9bea8-e5ac-4632-a292-a80a45831dff
  32. </details>
  33. ## News
  34. - **[2026-11]** v7.25.0: `autoloop_rebind_role` changes one role's engine, model or effort in a running autoloop
  35. run, without restarting it.
  36. - **[2026-11]** v7.24.0: the `opencode` engine runs OpenCode 2 as well as OpenCode 1.x, and GPT-6.2 Astra and Sol
  37. are registered.
  38. - **[2026-11]** v7.23.0: Antigravity sessions follow agy 1.5.9: every permission mode is passed explicitly, and
  39. Gemini 4 models are registered, with `agy-pro` on Gemini 4 Argon.
  40. - **[2026-11]** v7.22.0: Claude sessions can run under several Claude accounts and move to the next one when
  41. an account reaches its usage limit, continuing the same conversation.
  42. - **[2026-11]** v7.21.0: the plugin loads when OpenClaw installs it, with no native module to build.
  43. - **[2026-11]** v7.20.0: a fan-out agent can run as a goal, so several agents can each work a copy of a
  44. project until it is done, in parallel.
  45. - **[2026-11]** v7.19.0: a fan-out agent can run in a directory of its own, so several agents can write code
  46. in parallel, one copy each.
  47. - **[2026-11]** v7.18.0: `CLAWO_CLAUDE_GATEWAY=1` runs a `claude` session on any model your own
  48. Anthropic-compatible gateway serves, by its own name.
  49. - **[2026-11]** v7.17.0: an agent can resume a paused autoloop run itself (`autoloop_resume`), and a Codex goal
  50. that hits a model at capacity is resumed on its thread instead of failing.
  51. - **[2026-10]** v7.16.0: `clawo goal --engine claude` runs a goal on Claude Code's `/goal`, with the same
  52. checkpointing, resume and verification as a Codex goal.
  53. - **[2026-10]** [ResearchBoost-Agent](https://github.com/AI4AI-for-Anything/ResearchBoost-Agent), an autonomous agent
  54. that profiles, optimizes and validates AI training code, now runs its goal mode on Claw Orchestrator.
  55. - **[2026-10]** v7.14.0: any Agent Client Protocol agent runs as a session (`engine: 'acp'`), with presets for
  56. Kimi Code and dsh.
  57. - **[2026-10]** v7.13.0: `clawo goal` keeps a Codex `/goal` running for hours on one thread, reattaches it after a
  58. crash, and judges the result with your checks.
  59. - **[2026-09]** [Declare, Compile, Look](https://openreview.net/forum?id=Romd50Ce0w), built on Claw Orchestrator,
  60. was presented as a talk at the ECCV 2026 Workshop on Multimodal Digital Agents.
  61. - **[2026-08]** v6.0.0: the durable run kernel and the verification plane. A run survives its own process dying, and
  62. a run with an acceptance contract is judged by the runtime, not by the agent.
  63. ## Highlights
  64. - ✅ **Verdicts the agent cannot write**: acceptance contracts (commands, HTTP probes, screenshots, diff policy, file
  65. assertions) run by the runtime. A run ends `verified`, `refuted` or `unverified`, and editing the tests its contract
  66. runs refutes it ([guide](./skills/references/verification.md)).
  67. - 🔀 **Many vendors, one API**: Claude Code, Codex, Antigravity, Grok Build, OpenCode, any ACP agent and custom CLIs.
  68. `clawo fanout` asks several at once; `session_handoff` moves a live conversation to another engine in the same
  69. workspace ([guide](./skills/references/multi-engine.md)).
  70. - 💾 **Survives a crash**: every state transition is checkpointed; resume, retry, timeout, cancel and steer come from
  71. the kernel ([guide](./skills/references/workflow.md)).
  72. - 🎯 **Long goals**: `clawo goal` runs a Codex `/goal` unattended on one thread, resumes it after usage limits within a
  73. bound, optionally inside your own sandbox ([guide](./skills/references/workflow.md#goal)).
  74. - 💰 **Known cost**: one per-turn ledger across engines (`clawo runs`) and a `maxBudgetUsd` cap the runtime enforces on
  75. every engine ([guide](./skills/references/observability.md)).
  76. - 🔌 **Fits your tools**: a CLI, an MCP server, an ACP agent for editors, an OpenAI-compatible endpoint, a TypeScript
  77. SDK and an OpenClaw plugin ([ways to run](#ways-to-run)).
  78. claude-squad, Conductor and Vibe Kanban are interfaces for running agents side by side in worktrees; Claude Code
  79. subagents and Codex's parallel attempts stay within one vendor. Claw Orchestrator is a headless runtime underneath:
  80. several vendors behind one API, durable workflows, and a verification step the agents cannot edit. They can be used
  81. together.
  82. ## Quick Start
  83. Requires Node 22+ and at least one engine CLI installed and logged in (`claude`, `codex`, `agy`, `grok` or `opencode`).
  84. ```bash
  85. npm install -g @enderfga/claw-orchestrator
  86. clawo serve & # local runtime + dashboard on 127.0.0.1:18796
  87. ```
  88. ```bash
  89. clawo solve "The test fails. Fix price.js." --engine codex \
  90. --check "npm test" --check "node ../holdout/price.spec.mjs" --wait
  91. clawo verify <runId> # each check, why it failed, the files that changed
  92. clawo runs --since 1h # every turn on every engine: model, tokens, cost, verdict
  93. ```
  94. ```bash
  95. clawo fanout "How would you speed up this test suite?" --engines claude,codex,grok --synthesize --wait
  96. clawo goal --objective-file GOAL.md --check "npm run bench:check" --check-timeout 120 --wait
  97. ```
  98. The server writes an access token to `~/.openclaw/server-token` and the CLI reads it from there. To sign in the
  99. browser once, open `http://127.0.0.1:18796/login?token=<token>&redirect=/dash`. The token grants full control,
  100. including running commands as the server's user, so keep it private; see
  101. [Security model](./skills/references/getting-started.md#security-model) before exposing the server beyond localhost.
  102. - 🧪 **Acceptance contracts and evidence bundles** ([guide](./skills/references/verification.md))
  103. - 🔁 **Workflows, resume and the goal node** ([guide](./skills/references/workflow.md))
  104. - 🖥️ **Every CLI command** ([guide](./skills/references/cli.md))
  105. - 🧰 **All 80 tools** ([guide](./skills/references/tools.md))
  106. ## Ways to run
  107. | | For | Start here |
  108. | ------------------------------ | ----------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- |
  109. | **CLI and dashboard** | Terminals and scripts | `clawo serve`, then [`cli.md`](./skills/references/cli.md) |
  110. | **MCP server** | Claude Code, Claude Desktop, Cursor, Cline, Continue, Zed, Windsurf, Goose, Hermes Agent | `claude mcp add clawo -- clawo-mcp`, then [`mcp.md`](./skills/references/mcp.md) |
  111. | **ACP agent** | Zed, JetBrains, Neovim, Emacs, the VS Code ACP extension, dsh | `clawo acp`, then [`acp.md`](./skills/references/acp.md) |
  112. | **OpenAI-compatible endpoint** | Any OpenAI SDK client or webchat, without changing call sites | `POST /v1/chat/completions` ([`openai-compat.md`](./skills/references/openai-compat.md)) |
  113. | **TypeScript SDK** | Embedding the runtime in your own service | `import { SessionManager } from '@enderfga/claw-orchestrator'` (below) |
  114. | **OpenClaw plugin** | Every OpenClaw agent gets all 80 tools | `curl -fsSL https://raw.githubusercontent.com/Enderfga/claw-orchestrator/main/install.sh \| bash` |
  115. ```ts
  116. import { SessionManager } from '@enderfga/claw-orchestrator';
  117. const manager = new SessionManager();
  118. await manager.startSession({ name: 'fix-tests', engine: 'claude', cwd: '/project' });
  119. const result = await manager.sendMessage('fix-tests', 'Fix the failing tests');
  120. ```
  121. From an MCP host, ask for `workflow_start` with template `solve`, a task, and a contract such as
  122. `{"checks":[{"type":"command","cmd":"npm","args":["test"]}]}`. In an ACP editor, the model selector is grouped by
  123. engine, so one dropdown switches between Claude, Codex, Antigravity and Grok models mid-session.
  124. ## Engine compatibility
  125. | Engine | CLI | Tested version |
  126. | ----------- | ---------- | ------------------------------- |
  127. | Claude Code | `claude` | 2.1.327 |
  128. | Codex | `codex` | 0.169.1 |
  129. | Antigravity | `agy` | 1.5.10 |
  130. | Grok Build | `grok` | 1.0.88 |
  131. | OpenCode | `opencode` | 1.19.6, 2.3.2 |
  132. | ACP agent | any | Kimi Code 2.3.1, dsh 0.2.1-rc.2 |
  133. | Custom CLI | any | — |
  134. An agent that serves the Agent Client Protocol on stdio runs as
  135. [`engine: 'acp'`](./skills/references/multi-engine.md#acp-agents-engine-acp); any other coding CLI that runs as a
  136. subprocess can be wired up as a [custom engine](./skills/references/multi-engine.md#custom-engine-engine-custom).
  137. <details>
  138. <summary><b>How the table stays current</b></summary>
  139. The versions above are what the weekly sweep last ran. `scripts/sweep.ts` measures each core engine's installed,
  140. pinned and upstream version, diffs the flags the wrapper passes against the binary's `--help`, runs one live turn
  141. **through the real wrapper class**, smokes the ACP and MCP entry points, and checks `src/models.ts` against the
  142. OpenAI, Anthropic and xAI published price tables. It contains no model call, so the check that reports a wrapper as broken does not
  143. share the wrapper's failure modes.
  144. `scripts/sweep-workflow.json` wraps the same script as a durable run on this project's own kernel: verifier → router
  145. → an agent that drafts the alignment on a `sweep/<date>` branch → a human gate. A person reviews and merges; the
  146. orchestrator does not edit its own code unattended.
  147. </details>
  148. ## All features
  149. | Capability | What it does | Reference |
  150. | --------------------------- | ----------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------- |
  151. | **Persistent sessions** | Long-lived agents kept alive across requests, with context, tool, model and worktree control. | [`sessions.md`](./skills/references/sessions.md) |
  152. | **Multi-engine runtime** | One interface over the engines above, any ACP agent, and custom CLIs. | [`multi-engine.md`](./skills/references/multi-engine.md) |
  153. | **Session handoff** | Moves a live conversation to another engine or model in the same workspace; the old session keeps running. | [`sessions.md`](./skills/references/sessions.md) |
  154. | **Claude accounts** | Several Claude accounts, keys or gateways; a session moves to the next one at its usage limit and keeps its conversation. | [`multi-engine.md`](./skills/references/multi-engine.md#claude-accounts-and-rotation)|
  155. | **Verification plane** | Acceptance contracts the runtime runs itself, with an evidence bundle on disk; a run without a contract ends `unverified`. | [`verification.md`](./skills/references/verification.md) |
  156. | **Durable run kernel** | Declarative workflows over `agent`, `goal`, `fanout`, `council`, `verifier`, `human_gate`, `router` and more, checkpointed. | [`workflow.md`](./skills/references/workflow.md) |
  157. | **Fan-out** | One task across N engine/model agents in parallel, with an optional synthesis pass. | [`tools.md`](./skills/references/tools.md) |
  158. | **Council** | Parallel agents in isolated git worktrees, voting until they reach consensus. | [`council.md`](./skills/references/council.md) |
  159. | **Autoloop** | A Planner, Coder and Reviewer iterating on a workspace, each with its own engine and model; you chat with the Planner. | [`autoloop.md`](./skills/references/autoloop.md) |
  160. | **Ultraapp** | A short structured interview turned into a deployed web app at `localhost:19000/forge/<slug>/`. | [`ultraapp.md`](./skills/references/ultraapp.md) |
  161. | **ultracode** | `session_start({ ultracode: true })` lets Claude orchestrate a dynamic workflow of subagents. | [`tools.md`](./skills/references/tools.md) |
  162. | **Run ledger, spend caps** | Every turn on every engine in one JSONL ledger (`clawo runs`), and a `maxBudgetUsd` cap enforced on every engine. | [`observability.md`](./skills/references/observability.md) |
  163. | **OpenAI-compatible proxy** | `POST /v1/chat/completions` routed to persistent sessions on the engine the model name selects, streamed back in OpenAI shape. | [`openai-compat.md`](./skills/references/openai-compat.md) |
  164. | **Dashboard** | Autoloop, Council and Forge tabs with lifecycle controls and live event streams, and a Runs tab of durable workflow runs. | [`dashboard.md`](./skills/references/dashboard.md) |
  165. ## Built on Claw Orchestrator
  166. - [ResearchBoost-Agent](https://github.com/AI4AI-for-Anything/ResearchBoost-Agent): an autonomous agent that profiles,
  167. optimizes and validates AI training code; its long goal-mode runs use `clawo goal`
  168. ([blog](https://ydove0324.github.io/hx_resume/blog/researchboost/)).
  169. - [Declare, Compile, Look](https://openreview.net/forum?id=Romd50Ce0w): coordinate-free layout generation with
  170. vision-in-the-loop repair, ECCV 2026 Workshop on Multimodal Digital Agents (talk).
  171. Using Claw Orchestrator in a project or paper? Open a pull request to add it here.
  172. ## Citation
  173. If Claw Orchestrator helps your research, please cite its
  174. [technical report](https://doi.org/10.5281/zenodo.23874036):
  175. ```bibtex
  176. @techreport{fang2026claworchestrator,
  177. title = {Claw Orchestrator: A Unified Runtime for Coding Agent CLIs},
  178. author = {Fang, Guian},
  179. year = {2026},
  180. publisher = {Zenodo},
  181. doi = {10.5281/zenodo.23874036},
  182. url = {https://doi.org/10.5281/zenodo.23874036}
  183. }
  184. ```
  185. Research built on Claw Orchestrator:
  186. ```bibtex
  187. @misc{researchboost2026infra,
  188. author = {{ResearchBoost Team}},
  189. title = {{ResearchBoost}: {RSI} starts with Infra},
  190. year = {2026},
  191. url = {https://ydove0324.github.io/hx_resume/blog/researchboost/}
  192. }
  193. @inproceedings{fang2026declare,
  194. title = {Declare, Compile, Look: Coordinate-Free Layout Generation with Vision-in-the-Loop Repair},
  195. author = {Fang, Guian and Liu, Mengsha and Shou, Mike Zheng},
  196. booktitle = {ECCV 2026 Multimodal Digital Agents Workshop},
  197. year = {2026},
  198. url = {https://openreview.net/forum?id=Romd50Ce0w}
  199. }
  200. ```
  201. ## Contributing
  202. See [`CONTRIBUTING.md`](./CONTRIBUTING.md). Run `npm run build && npm run lint && npm run format:check && npm run test`
  203. before submitting.
  204. ## License
  205. MIT, see [`LICENSE`](./LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.