← 返回 Skills 市场
lludlow

Clawctl

作者 lludlow · GitHub ↗ · v0.1.0
cross-platform ✓ 安全检测通过
836
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install clawctl
功能描述
Coordination layer for OpenClaw agent fleets (tasks, messaging, activity feed, dashboard).
使用说明 (SKILL.md)

Setup

clawctl init                        # create the database
export CLAW_AGENT=your-name         # set identity (falls back to $USER with warning)
export CLAW_DB=~/.openclaw/clawctl.db  # optional, this is the default

Operational Rhythm

Follow this pattern every session:

  1. clawctl checkin — register presence, see unread count
  2. clawctl inbox --unread — read messages before picking up work
  3. clawctl next — find highest-priority actionable task (or clawctl list --mine)
  4. clawctl claim \x3Cid> then clawctl start \x3Cid> — take ownership and begin
  5. clawctl msg \x3Cagent> "update" --task \x3Cid> — coordinate during work
  6. clawctl done \x3Cid> -m "what I did" then clawctl next — complete and move on

Only claim tasks assigned to you or matching your role. Completing an already-done task is a safe no-op.

Decision Tree

Situation Command
New task clawctl add "Subject" -d "Details"
Find work clawctl next then clawctl claim \x3Cid>
Blocked clawctl block \x3Cid> --by \x3Cblocker-id> and notify via clawctl msg
Finished clawctl done \x3Cid> -m "Result"
Hand off clawctl msg \x3Cagent> "Ready for you" --task \x3Cid> --type handoff
Ready for review clawctl review \x3Cid>
Catch up clawctl feed --last 20 or clawctl summary
Link artifacts Add --meta '{"note":"path/to/file"}' to done, claim, start, or block

Task Statuses

pending → claimed → in_progress → done
                  ↘ blocked ↗    ↘ cancelled
                  ↘ review  ↗

list excludes done/cancelled by default. Use --all for history (newest first).

Commands

Tasks

Command Description
add SUBJECT Create task. -d desc, -p 0|1|2 priority, --for AGENT assign, --parent ID
list Active tasks. --mine, --status STATUS, --owner AGENT, --all
next Highest-priority actionable task for current agent
claim ID Claim task. --force overrides ownership, --meta JSON
start ID Begin work (in_progress). --meta JSON
done ID Complete. -m note, --force, --meta JSON
review ID Mark ready for review. --meta JSON
cancel ID Cancel task. --meta JSON
block ID --by OTHER Mark blocked. --meta JSON
board Kanban board grouped by status

Messages

Command Description
msg AGENT BODY Send message. --task ID, --type TYPE (comment, status, handoff, question, answer, alert)
broadcast BODY Alert all agents
inbox Read messages. --unread for unread only

Fleet

Command Description
checkin Heartbeat — update presence, report unread count
register NAME Register agent. --role TEXT
fleet All agents with status and current task
whoami Identity, role, and DB path

Monitoring

Command Description
feed Activity log. --last N, --agent NAME, --meta
summary Fleet overview with counts and recent events
dashboard Web UI. --port INT, --stop, --verbose
安全使用建议
This skill appears to be what it says: a local coordination CLI + optional Flask dashboard backed by SQLite. Before installing or running it, consider the following: - Install method: the registry metadata omits an install step even though a Python package (pyproject.toml) is included. You should install into a virtualenv (pip install . or pip install -e . for development) rather than running unknown files globally. - Database safety: run initial tests with a temporary DB to avoid touching your real data (example: CLAW_DB=/tmp/test.db clawctl init). The README/CLAUDE.md recommend this. - Dashboard token and exposure: the dashboard writes a persistent token to ~/.openclaw/.clawctl-token and, by default, binds to 0.0.0.0. If you run the dashboard on a multi-user or public-facing host, ensure the token file is protected (filesystem permissions) and the port is not publicly accessible, or bind to localhost instead. - Environment vars: the code reads CLAW_AGENT and CLAW_DB (with sensible defaults). The skill metadata did not declare these env vars — be aware they influence behavior and identity fallback to $USER will occur with a one-time warning. - Review the code if you need stricter guarantees: SQL is parameterized and DB access patterns look race-aware; there are no obvious network calls to external services or obfuscated code. If you require tighter isolation, run inside a container or VM and use a temp DB. If you want, I can extract the full list of CLI commands or point out exact lines where the dashboard writes the token and binds to the network so you can audit or modify them before use.
功能分析
Type: OpenClaw Skill Name: clawctl Version: 0.1.0 The OpenClaw AgentSkills skill bundle 'clawctl' is a coordination layer for agent fleets, utilizing a local SQLite database and an optional Flask web dashboard. The code demonstrates strong security practices, including parameterized SQL queries in `clawctl/db.py` to prevent SQL injection, HTML escaping in `dashboard/index.html` to mitigate XSS, and controlled execution of the dashboard server via `subprocess.Popen` in `clawctl/cli.py` with fixed arguments. Data is stored locally in user-owned directories (`~/.openclaw/clawctl.db`, `~/.openclaw/.clawctl-token`), and there is no evidence of data exfiltration, malicious execution, persistence mechanisms, or prompt injection attempts against the agent. All functionalities are aligned with the stated purpose of a local coordination tool.
能力评估
Purpose & Capability
Name/description match the code and SKILL.md: CLI (clawctl), DB layer, and Flask dashboard implement task board, messaging, activity feed, and a web UI. The declared required binary (clawctl) is appropriate. Minor inconsistency: registry/install metadata treats this as instruction-only, but the package source (pyproject, Python package files) is included — not dangerous but the packaging/installation story is unclear.
Instruction Scope
SKILL.md instructs running clawctl CLI commands and setting CLAW_AGENT/CLAW_DB; runtime instructions stay within the coordination/dashboard scope. Commands allow including JSON metadata that may reference local file paths (e.g. --meta '{"note":"path/to/file"}'), but the skill does not instruct the agent to read arbitrary system files or exfiltrate data. Dashboard persists a token file under ~/.openclaw/.clawctl-token — expected for local web UI auth.
Install Mechanism
No network install or remote download specs are present; source includes a standard pyproject.toml with only Click and Flask dependencies. This is low risk compared with remote fetch installs. The absence of an install spec means users must install the package (pip) or otherwise provide the clawctl binary.
Credentials
The skill does not request any external credentials and uses only CLAW_AGENT and CLAW_DB (both optional; defaults exist). Those env vars are referenced throughout the code and SKILL.md but are not listed in requires.env metadata — this mismatch is minor but worth documenting. No unrelated secrets or cloud credentials are requested.
Persistence & Privilege
always:false and no skill-wide privileged flags are used. The dashboard creates and persists a local token file at ~/.openclaw/.clawctl-token and runs a Flask server bound to 0.0.0.0 by default (intended for LAN/Tailscale access). That network exposure is expected for a dashboard but requires the user to be careful about where they run it and who can access the token file or the listening port.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install clawctl
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /clawctl 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.0
Initial release of clawctl: coordination tool for OpenClaw agent fleets. - Provides task management (add, claim, start, complete, block, review, cancel). - Built-in messaging between agents, including inbox, broadcast, and task-specific messages. - Fleet operations: agent registration, presence check-in, whoami, and live dashboard. - Activity monitoring with feed, summary, and board views. - Simple operational rhythm and command decision tree for efficient workflow.
元数据
Slug clawctl
版本 0.1.0
许可证
累计安装 4
当前安装数 0
历史版本数 1
常见问题

Clawctl 是什么?

Coordination layer for OpenClaw agent fleets (tasks, messaging, activity feed, dashboard). 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 836 次。

如何安装 Clawctl?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install clawctl」即可一键安装,无需额外配置。

Clawctl 是免费的吗?

是的,Clawctl 完全免费(开源免费),可自由下载、安装和使用。

Clawctl 支持哪些平台?

Clawctl 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Clawctl?

由 lludlow(@lludlow)开发并维护,当前版本 v0.1.0。

💬 留言讨论