@teamclaws/teamclaw
v2026.4.3-2
Published
OpenClaw virtual software team orchestration plugin
Downloads
911
Maintainers
Readme
@teamclaws/teamclaw
TeamClaw is an OpenClaw plugin that turns multiple roles into a collaborative virtual software team.
It supports:
controller/workermodes- externally registered workers
- adaptive kickoff planning for medium/complex work, with per-role assessments before task creation
- clarifications, workspace browsing, controller UI visibility, and a matching desktop client
- Git-based collaboration
- on-demand worker provisioning with
process,docker, andkubernetes
Install
For the easiest guided setup, run:
npx -y @teamclaws/teamclaw installThe guided installer automatically uses OpenClaw's break-glass install flag for TeamClaw, because TeamClaw legitimately needs host orchestration capabilities that the built-in scanner marks as critical.
This installer can:
- install/update the TeamClaw plugin in OpenClaw
- detect your local
openclaw.json - let you choose the installation mode
- support dedicated worker-only installs with
--install-mode worker - let you choose a model from the models already defined in OpenClaw
- let you choose the OpenClaw workspace directory
- default TeamClaw to a dedicated
teamclawagent/workspace instead of reusingmain - copy the effective host model into the dedicated TeamClaw agent config when independent mode is used
- bootstrap TeamClaw auth profiles from the host OpenClaw auth store when available
- prefill Docker/Kubernetes provisioning with the published TeamClaw runtime image
- prefill Docker workspace persistence with a named volume and Kubernetes persistence with a PVC name
By default, guided install now uses an independent TeamClaw agent/workspace layout:
agent:teamclaw:*sessions- sibling workspace such as
~/.openclaw/workspace-teamclaw
In practice, this means TeamClaw keeps its own agent state, workspace, and auth-profile copy instead of reusing the host main agent. Use this default unless you explicitly need legacy compatibility with a shared main workspace.
For advanced compatibility only, you can force the legacy shared-main layout:
npx -y @teamclaws/teamclaw install --agent-mode mainIf you only want to install the plugin manually, use:
openclaw plugins install --dangerously-force-unsafe-install @teamclaws/teamclawIf you want to force the ClawHub package path, use:
openclaw plugins install --dangerously-force-unsafe-install clawhub:@teamclaws/teamclawThen enable and configure TeamClaw in your openclaw.json.
The published TeamClaw runtime image also preinstalls the clawhub CLI, so containerized workers can discover and install skills from ClawHub without an extra bootstrap step.
Maintainer release flow
For maintainers, TeamClaw now uses two release paths:
- npm package — published by GitHub Actions via
.github/workflows/teamclaw-plugin-npm-release.yml - ClawHub code plugin + bundled skills — published manually from the CLI
The npm publish workflow runs automatically when:
- a tag matching
v*is pushed - or relevant package files change on
main(src/**,.github/workflows/teamclaw-plugin-npm-release.yml,scripts/sync-teamclaw-plugin-manifest.mjs,scripts/teamclaw-package-check.mjs,scripts/teamclaw-npm-publish.sh)
It also supports workflow_dispatch for a specific commit SHA on main.
Before the first real npm publish, GitHub must already be configured for npm trusted publishing with the npm-release environment and this workflow file.
Before either release path, sync and validate the generated plugin manifest:
npm --prefix src run manifest:sync
npm --prefix src run release:checkPreview the ClawHub release commands without publishing:
npm --prefix src run release:clawhub:dry-runWhen ready to publish to ClawHub, log in first and then run:
clawhub login
bash scripts/teamclaw-clawhub-release.sh --publishThe ClawHub release helper publishes the @teamclaws/teamclaw code plugin from src/ and then publishes only the bundled skills under src/skills/. Update each skill's frontmatter version: in src/skills/*/SKILL.md before a real publish if that skill changed.
Recommended first setup
For a first-time setup, the safest path is:
- Start with single-host
processprovisioning - Validate the workflow with a small smoke-test task
- Expand to external workers, Docker, or Kubernetes after the basics are working
When on-demand provisioning is enabled, TeamClaw now treats controller startup as a readiness phase, not just a process-up check. During that warm-up, /api/v1/health returns a non-OK status until the controller has verified writable runtime paths and successfully brought the configured startup worker roles online. If workerProvisioningRoles is empty, startup readiness defaults to waiting for a warm developer worker.
Documentation
- Website: https://topcheer.github.io/teamclaw/
- Installation guide: https://github.com/topcheer/teamclaw/blob/main/INSTALL.md
- Repository overview: https://github.com/topcheer/teamclaw/blob/main/README.md
- Design notes: https://github.com/topcheer/teamclaw/blob/main/DESIGN.md
