← 返回 Skills 市场
lawliet-ai

Meta-Router

作者 Lawliet-ai · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ⚠ suspicious
275
总下载
1
收藏
1
当前安装
2
版本数
在 OpenClaw 中安装
/install meta-router
功能描述
Automatically indexes and routes user intents to installed OpenClaw skills by scanning SKILL.md files and dynamically injecting relevant skill contexts.
安全使用建议
This skill's purpose (indexing and routing other skills) is reasonable, but the runtime instructions ask the agent to create and silently maintain a hidden index file and to perform automatic background scans of your ~/.openclaw/skills directory—yet the registry entry did not declare any config paths or persistent state. Before installing: 1) ask the publisher to clarify the exact index path (is it ~/.openclaw/.meta_index.json or 'root' filesystem?), why persistent storage is needed, and why it wasn't declared; 2) request source code or a homepage so you can audit the implementation; 3) insist that mounting/auto-reindex require explicit user consent or at least visible logging (avoid silent background writes); 4) consider running it in a disposable or isolated agent environment first to see what files it creates; and 5) if you cannot verify provenance and behavior, do not install or disable autonomous invocation and background indexing.
功能分析
Type: OpenClaw Skill Name: meta-router Version: 1.0.1 The Meta-Router bundle is a utility designed to optimize OpenClaw performance by managing how other skills are indexed and loaded into the agent's context. It instructs the agent to maintain a local index file (`.meta_index.json`) and perform directory scans of the `~/.openclaw/skills/` path to ensure efficient routing. While the instructions emphasize 'stealth' and 'silence' for background operations, these are clearly aimed at reducing conversational noise and token usage rather than hiding malicious activity. The scope of file access is limited to the OpenClaw environment, and there are no indicators of data exfiltration or unauthorized remote execution.
能力评估
Purpose & Capability
The skill claims to index and route installed skills, which aligns with instructions to read SKILL.md files. However, the runtime docs require creating and reading a hidden index file ('.meta_index.json') under the user's environment (AGENT.md references ~/.openclaw/.meta_index.json), yet the registry metadata declares no required config paths or persistent storage. The source/homepage is also missing, making provenance unclear.
Instruction Scope
SKILL.md/AGENT.md instruct the agent to perform background, silent scans and to create/maintain a hidden persistent index; trigger immediate atomic re-indexes on directory hash changes; and mount skills automatically (including bypass via '!' shortcuts). These behaviors go beyond a simple read-only index lookup: they require write access, background processes, and automatic actions that the user may not expect. The docs also ambiguously refer to 'root directory' vs ~/.openclaw, which is inconsistent.
Install Mechanism
There is no install spec and no code files (instruction-only), which reduces supply-chain risks. However, the skill explicitly requires creating a persistent hidden file on disk and performing automatic indexing—disk writes are part of the runtime instructions even though nothing is installed by package managers.
Credentials
The skill declares no required environment variables or config paths, yet the instructions expect read/write access to ~/.openclaw (or a 'root' index) and to monitor ~/.openclaw/skills/. This mismatch (undeclared filesystem access) is disproportionate and should have been declared explicitly so users can consent to persistent state and directory monitoring.
Persistence & Privilege
The skill requests persistent presence via a hidden index file, silent background indexing, and automatic immediate re-index triggers. Although 'always' is false, autonomous invocation is enabled by default; combined with silent filesystem writes and automatic mounting on shortcuts, this raises the blast radius if the skill behaves unexpectedly. It does not request system-wide configuration changes, but it does create persistent state without declaring it.
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 meta-router
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /meta-router 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
v1.0.1
- Hardened core logic: Router now blocks all task processing until context-skill alignment is confirmed. - State integrity improved: Enforced atomic rescan on skill changes and compressed descriptions to prevent token bloat. - Routing updated: Explicit commands and shortcut IDs bypass analysis. No more than two skills can be mounted at once unless Hive-Commander is used. - Mounting made volatile: All metadata is pruned immediately at the end of a workflow. - Operational efficiency: Enforced sub-200ms scanning, MRU eviction if the index exceeds 4KB, and continued silent background tasks except for I/O errors.
v1.0.0
Stop Token Bloat. Start Routing. ⚡ As you install more skills, your Agent becomes slower and more expensive because it prepends every skill definition to every prompt. Meta-Router changes the game by introducing a two-phase "Scan & Route" logic.
元数据
Slug meta-router
版本 1.0.1
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 2
常见问题

Meta-Router 是什么?

Automatically indexes and routes user intents to installed OpenClaw skills by scanning SKILL.md files and dynamically injecting relevant skill contexts. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 275 次。

如何安装 Meta-Router?

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

Meta-Router 是免费的吗?

是的,Meta-Router 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Meta-Router 支持哪些平台?

Meta-Router 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 Meta-Router?

由 Lawliet-ai(@lawliet-ai)开发并维护,当前版本 v1.0.1。

💬 留言讨论