菜单
Use this skill when you are a worker operating inside a CSGClaw task execution room. Only act on explicit task dispatch messages, claim the dispatched task, report blocked/completed/failed status through the team CLI, and keep all execution aligned with server task state.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run the complete three-stage CSGClaw structured-output demo when the user explicitly invokes $csgclaw-interactive-output-demo. Exercise resource links, Codex-style options, recommended and Unicode labels, freeform alternatives, secret input, and agent-directed automatic continuation.
Mandatory skill for provisioning any new CSGClaw agent-backed participant or worker. Use immediately when the user asks to create, add, set up, or provision an agent, robot, worker, or user-facing "bot" (including GitLab, frontend, backend, QA, or other specialized workers), when dispatch needs a missing worker, or when asking which hub template fits. Always template list + match + template get + participant create --type agent --bind create --from-template with --env for secrets. Never create a CSGClaw worker with --type agent unless it binds a real Agent. Never run participant create --bind create without --from-template for a new worker. Do NOT use for task dispatch to existing workers.
Default manager skill for multi-worker CSGClaw orchestration. Create teams and tasks, plan and start work through `csgclaw-cli team`, and keep dispatch, claim, dependencies, and status on server task state.
Configure and troubleshoot CSGClaw Feishu/Lark channel credentials for manager or worker agents. Use when the Manager needs to generate a Feishu bot app creation URL or QR code, collect App ID/App Secret through registration, bind Feishu participants through `csgclaw-cli participant bind`, recreate workers, or debug Feishu messages not reaching CSGClaw workers.
Search, inspect, and install ClawHub-compatible skills with csgclaw-cli skill. Use when the user asks to find skills on opencsg or clawhub.ai, list versions, or install into this sandbox workspace/skills directory.
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations.
| name | agent-teams |
| description | Use this skill when you are a worker operating inside a CSGClaw task execution room. Only act on explicit task dispatch messages, claim the dispatched task, report blocked/completed/failed status through the team CLI, and keep all execution aligned with server task state. |
Use this skill when the manager is coordinating work through CSGClaw team tasks.
Only begin work after an explicit dispatch message in the task execution room:
@<worker_display_name> <manager_display_name> dispatched <task_id> to <worker_display_name>. Claim: csgclaw-cli team task claim --team <team_id> --task <task_id> --participant-id <worker_participant_id>
The claim command uses participant IDs, for example u-frontend-dev. Rendered mentions may show only the display name or handle, for example @frontend-dev; do not use the display name as the CLI participant ID unless the claim command shows it.
Do not use legacy --bot-id for team commands. If csgclaw-cli team task claim --help does not show --participant-id, the sandbox has a stale csgclaw-cli; report that blocker and stop instead of retrying with --bot-id.
Ignore team setup and planning events, including messages such as enabled team, created task, created tasks, completed planning for task, and created execution room. Those messages are not permission to start work.
<team_id> and <task_id> from the dispatch messageblocked, completed, or failedDo not self-assign work. When the room dispatch message includes a task id, always use --task. Use claim-next only when no task id was provided.
Never infer team_id from the room id. A room id such as room-178... is not a team id. Use the --team <team_id> value shown in the dispatch message. If the only available value starts with room-, stop and report blocked instead of continuing.
Claim the dispatched task:
csgclaw-cli team task claim --team <team_id> --task <task_id> --participant-id <worker_participant_id>
If the dispatch did not include a task id, claim the next available task:
csgclaw-cli team task claim-next --team <team_id> --participant-id <worker_participant_id>
After claiming, confirm the response status is in_progress for the same task before doing the work.
Report a completed task:
csgclaw-cli team task update --team <team_id> --task <task_id> --actor-id <worker_participant_id> --status completed --result "<summary>"
The task is not complete until this CLI status update succeeds. Sending a normal room message with the result is useful, but it does not update task state. Do not update tracker files or rely on manual handoff; task completion is recorded only through the team CLI.
Report a blocked task:
csgclaw-cli team task update --team <team_id> --task <task_id> --actor-id <worker_participant_id> --status blocked --reason "<why blocked>"
Report a failed task:
csgclaw-cli team task update --team <team_id> --task <task_id> --actor-id <worker_participant_id> --status failed --error "<failure>"
~/.openclaw/workspace/projects/{name} as the shared project directory. Check there first before creating a new project path.~/.openclaw/workspace/projects/{name} tree so managers and workers can inspect the same content.<team_id>/<task_id> that you claimed; do not finish with only a chat message.blocked and explain the reason clearly.in_progress.