diff --git a/tool-submitters.json b/tool-submitters.json index 5d20a9a..f21296b 100644 --- a/tool-submitters.json +++ b/tool-submitters.json @@ -43,6 +43,7 @@ "one": "shawnbure", "onecli": "bradvin", "openbrowser": "bradvin", + "openrig": "foo-bender", "orthogonal": "bradvin", "pangolinfo-amazon-data-mcp": "Pangolin-spg", "paperclip": "bradvin", diff --git a/tools/openrig.md b/tools/openrig.md new file mode 100644 index 0000000..a7ff8b3 --- /dev/null +++ b/tools/openrig.md @@ -0,0 +1,75 @@ +--- +slug: "openrig" +name: "OpenRig" +description: "Local control plane for persistent Claude Code and Codex teams with durable work ownership, review, proof, and restore state" +agentSummary: "OpenRig lets a lead coding agent route work to named Claude Code and Codex seats without a person relaying messages between terminals. Its local control plane retains task ownership, proof, transcripts, and restore state, while the shipped starter separates implementation from an independent check before a person decides whether to publish the change." +seoTitle: "OpenRig Persistent Teams for Claude Code and Codex" +seoDescription: "Run named Claude Code and Codex seats as a local team with durable task ownership, independent checks, proof records, and explicit restore outcomes." +category: "orchestrators" +tags: + - "coding-agents" + - "multi-agent" + - "orchestration" + - "local-first" + - "tmux" +websiteUrl: "https://openrig.dev" +githubUrl: "https://github.com/mvschwarz/openrig" +pricing: "open-source" +classification: "agent-native" +entityType: "software-application" +developerName: "Mike Schwarz" +docsUrl: "https://github.com/mvschwarz/openrig/blob/v0.6.4/docs/reference/getting-started.md" +licenseUrl: "https://github.com/mvschwarz/openrig/blob/v0.6.4/LICENSE" +interfaces: + - "CLI" + - "terminal UI" + - "MCP" +deploymentModes: + - "local self-hosted" +evidenceSources: + - title: "OpenRig product page" + url: "https://openrig.dev" + claim: "OpenRig documents persistent coding-agent teams, direct specialist messaging, durable task ownership, and a terminal UI for seeing team structure and work." + accessedAt: "2026-10-02" + sourceType: "official-product-page" + - title: "OpenRig v0.6.4 README" + url: "https://github.com/mvschwarz/openrig/blob/v0.6.4/README.md" + claim: "The tagged README documents the local CLI and TUI, named Claude Code and Codex seats, queue-backed work, reviewed starter workflow, snapshot and restore behavior, supported operating systems, and runtime prerequisites." + accessedAt: "2026-10-02" + sourceType: "official-documentation" + - title: "OpenRig v0.6.4 getting-started guide" + url: "https://github.com/mvschwarz/openrig/blob/v0.6.4/docs/reference/getting-started.md" + claim: "The tagged guide defines an outcome owner and independent checker, durable queue ownership, review of the exact candidate, and the boundary between OpenRig orchestration and provider permission or sandbox controls." + accessedAt: "2026-10-02" + sourceType: "official-documentation" + - title: "OpenRig v0.6.4 release" + url: "https://github.com/mvschwarz/openrig/releases/tag/v0.6.4" + claim: "The stable release identifies CLI and TUI as supported interfaces, places the web UI in maintenance mode, and records current restore, sandbox, reboot, and platform limitations." + accessedAt: "2026-10-02" + sourceType: "official-product-announcement" + - title: "OpenRig v0.6.4 Apache 2.0 license" + url: "https://github.com/mvschwarz/openrig/blob/v0.6.4/LICENSE" + claim: "The reviewed OpenRig release is licensed under Apache License 2.0 and identifies Mike Schwarz as the copyright holder." + accessedAt: "2026-10-02" + sourceType: "official-license" +verificationLevel: "documentation-reviewed" +reviewedBy: "foo-bender" +reviewedAt: "2026-10-03" +classificationRationaleMd: "OpenRig is built around coding agents as persistent, named workers: a lead seat routes work to specialist Claude Code and Codex seats while the control plane retains team topology, task ownership, communication, proof, and recovery state." +inclusionRationaleMd: "OpenRig owns a material lifecycle across coding-agent runs. It assigns queue-backed work to named seats, runs workers in separate tmux sessions, retains shared state and evidence, and routes a candidate from an owner to an independent checker before a person decides whether to publish it." +bestForMd: "Developers who already use Claude Code or Codex and want a local, persistent team with explicit roles, durable work ownership, direct cross-harness communication, independent checks, and recoverable seat identities." +notBestForMd: "Teams seeking a managed cloud service, strong multi-tenant security isolation, native Windows support, or automatic code merging and deployment." +limitationsMd: "The reviewed v0.6.4 release requires Node.js 22 or 24 and tmux on macOS or Linux; Apple silicon Macs should use Node.js 22, and WSL2 is untested. Claude Code or Codex accounts must be authenticated separately, and provider model-usage costs still apply. OpenRig provides operational separation, not a security boundary; starter seats share a working directory. Daemon startup and managed launches write provider hooks, trust records and settings, including pre-trusting workspaces; OPENRIG_HOME does not isolate those provider files. Back up relevant settings because there is no complete preservation or rollback guarantee, and rig setup --dry-run does not preview every later effect. Native provider permissions and sandboxes govern execution, while broader command allowances require an explicit choice. Keep the daemon on loopback or a trusted private network, not the public internet. Known release limits include a restore check that can incorrectly flag a working Claude seat, the default Codex sandbox blocking daemon access, and no automatic daemon restart after reboot." +unknownsMd: "This review did not install or run OpenRig. Pi appears in current marketing and repository source but not in the tagged v0.6.4 first-use path, so it is not claimed here. Slack, multi-host operation, restore behavior, and cross-seat handoffs were not independently tested. The website documentation still identifies version 0.5.14, so release-specific claims are pinned to v0.6.4 sources." +--- + +OpenRig is a local control plane for organizing ordinary Claude Code and Codex sessions into a persistent team. It gives each role a stable seat and address, tracks owned work in a durable queue, and keeps project, proof, transcript, and restore state outside any one agent's context window. + +Unlike a desktop workspace, ticket runner, company-governance layer, or credential gateway, OpenRig focuses on the topology of a coding-agent team. Agents communicate through their terminal sessions, route work to named specialists, and return reviewed evidence while the person remains responsible for deciding what should be published or landed. + +## So agents can... + +- Assign and hand off durable queue work to named Claude Code or Codex seats. +- Ask specialists directly and retain communication, task, project, and proof state beyond one session. +- Route an implementation to an independent checker and record the result without a person relaying messages between terminals. +- Restore the same team addresses after interruption and report whether each seat resumed, rebuilt, started fresh, or failed.