← 返回 Skills 市场
OpenClaw Manager
作者
ZephyrChen0754
· GitHub ↗
· v0.3.2
· MIT-0
205
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install humanclaw-made-openclaw-manager
功能描述
Install or operate a standalone local OpenClaw manager skill that adds shadow-first thread observation, durable session/run state, a loopback-only sidecar, a...
使用说明 (SKILL.md)
OpenClaw Manager
Use this skill when the task is to operate, inspect, or extend the local OpenClaw Manager control plane.
What this skill owns
- local
session / run / event / checkpoint / attentionstate - shadow-first
thread_shadowobservation and promotion queue - append-only
events.jsonlandskill_traces.jsonl - local snapshot export
- connector normalization for Telegram, WeCom, Email, and GitHub
- capability graph and anonymized fact export
- standalone sidecar bootstrap and local command surface
- loopback-only sidecar by default
- consent-gated sidecar autostart
Entry points
- bootstrap runtime:
src/skill/bootstrap.ts - local sidecar API:
src/api/server.ts - command registry:
src/skill/commands.ts - connector registry:
src/connectors/registry.ts - capability graph:
src/telemetry/capability-graph.ts
References
- architecture:
docs/architecture.md - session model:
docs/session-model.md - event schema:
docs/event-schema.md - connector protocol:
docs/connector-protocol.md - capability facts:
docs/capability-facts.md - security model:
SECURITY.md
安全使用建议
Before installing: 1) Treat this as a local Node.js service — ensure you want a background sidecar and local durable state under ~/.openclaw/skills/manager. 2) Verify you have Node.js >=20 and npm available (the installer requires these). 3) Inspect the code (bootstrap, sidecar-launcher, connectors) yourself or run the provided security smoke tests (node scripts/security-smoke.cjs and node scripts/smoke-test.cjs) in an isolated environment. 4) Confirm the npm registry and lockfile are acceptable to you (installer enforces official registry). 5) Note that connectors are opt-in but will require secrets if enabled; keep those secrets in private config and avoid committing them. 6) Only grant the autostart consent if you are comfortable the launcher will spawn a local Node process; you can run the sidecar manually instead. 7) If you lack the ability to audit the code, run the skill in a sandboxed user account or VM and back up any sensitive local data first.
功能分析
Type: OpenClaw Skill
Name: humanclaw-made-openclaw-manager
Version: 0.3.2
The OpenClaw Manager is a local control plane designed to manage agent session state, thread observation (shadowing), and capability reporting. The bundle implements a local sidecar architecture using Express, which is strictly configured to bind to loopback (127.0.0.1) by default and requires explicit user consent for autostart functionality (src/skill/bootstrap.ts). Security-focused features are present, such as a security smoke test (scripts/security-smoke.cjs) that validates network boundaries and an installation script that enforces the official npm registry to prevent dependency confusion. No evidence of data exfiltration, unauthorized remote access, or malicious prompt injection was found; the tool's high-privilege actions (filesystem access and local process spawning) are well-documented and aligned with its stated purpose.
能力评估
Purpose & Capability
The code, docs, and SKILL.md consistently describe a local Node.js 'OpenClaw Manager' that hosts a loopback sidecar, filesystem-first state, connector adapters, and telemetry exports. That behavior matches the name and description. However, the registry metadata claims 'required binaries: none' and 'required env vars: none', which is inaccurate: the project requires Node.js (node, npm) and uses many OPENCLAW_MANAGER_* configuration envs and a CODEX_HOME for optional install. This metadata mismatch is an incoherence that should be addressed.
Instruction Scope
The SKILL.md and included scripts limit network activity to loopback by default and document connector opt-in behavior. Runtime instructions read local manager settings, write durable state under a documented directory, run local health checks, and may spawn a local Node process to start the sidecar (consent-gated). They do not, by default, upload raw transcripts or phone home. Connectors (Telegram/WeCom/Email/GitHub) are present but are explicitly opt-in and require additional configuration/secrets to make external requests.
Install Mechanism
There is no registry install spec in the metadata, but the repo contains an installer script (scripts/install.sh) that runs 'npm ci' and 'npm run build'. The installer checks the npm registry and the lockfile and uses the official npm registry; that is reasonable but still involves fetching packages from npm. The presence of a full package-lock and many source files increases the surface area vs. a pure instruction-only skill. The installer also prompts for autostart consent and can copy files into $CODEX_HOME. This is moderate-risk but expected for a Node-based local sidecar.
Credentials
Registry metadata lists no required env vars, yet the code and docs read many configuration env vars (OPENCLAW_MANAGER_STATE_ROOT, OPENCLAW_MANAGER_BIND_HOST, OPENCLAW_MANAGER_SIDECAR_URL, OPENCLAW_MANAGER_ALLOW_REMOTE_SIDECAR, OPENCLAW_MANAGER_NO_AUTOSTART, OPENCLAW_MANAGER_SERVER_PROCESS, PORT, and optionally CODEX_HOME). While these are configuration values (not secrets) and are justified by the project's behavior, the absence of these in the declared metadata is an incoherence. Connectors can require secrets when enabled, which is documented; those secrets are local but should be treated with care.
Persistence & Privilege
The skill persists local state under a documented default (~/.openclaw/skills/manager) and stores a one-time autostart consent flag there. It can copy itself into $CODEX_HOME when asked. It does not request 'always: true' and does not claim to modify other skills. Autostart is consent-gated and disabled by default. These privileges are proportionate to a local manager, though autostart and process-spawning increase risk and should be explicitly consented to.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install humanclaw-made-openclaw-manager - 安装完成后,直接呼叫该 Skill 的名称或使用
/humanclaw-made-openclaw-manager触发 - 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
v0.3.2
Security hardening release: loopback-only default sidecar binding, consent-gated autostart, official npm registry lockfile cleanup, security smoke checks, and clearer local-sidecar disclosure.
v0.3.1
Public-ready v1. Synced skill metadata to 0.3.1, cleaned the README, added the fastest-path install flow, install modes, troubleshooting notes, and a 3-minute quickstart.
元数据
常见问题
OpenClaw Manager 是什么?
Install or operate a standalone local OpenClaw manager skill that adds shadow-first thread observation, durable session/run state, a loopback-only sidecar, a... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 205 次。
如何安装 OpenClaw Manager?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install humanclaw-made-openclaw-manager」即可一键安装,无需额外配置。
OpenClaw Manager 是免费的吗?
是的,OpenClaw Manager 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
OpenClaw Manager 支持哪些平台?
OpenClaw Manager 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 OpenClaw Manager?
由 ZephyrChen0754(@zephyrchen0754)开发并维护,当前版本 v0.3.2。
推荐 Skills