← 返回 Skills 市场
Metacognition
作者
JoeyCacciatore3
· GitHub ↗
· v1.0.0
· MIT-0
89
总下载
1
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install nate-metacognition
功能描述
Self-reflection engine for AI agents. Extracts patterns from session transcripts into a weighted graph with Hebbian learning and time decay. Compiles a token...
安全使用建议
This skill appears to implement the claimed metacognition functionality, but the code contradicts its own security statements: metacognition.py calls 'curl' via subprocess to reach the embeddings endpoint and does not validate EMBEDDINGS_URL to localhost. That means if EMBEDDINGS_URL is set to a remote server, the skill could transmit text (including session content) to that server. Before installing, either: 1) inspect metacognition.py and remove/replace subprocess+curl with a validated local-only HTTP client (e.g., Python urllib with an allowlist of 127.0.0.1/::1), 2) ensure the runtime environment cannot reach remote addresses (network sandboxing), or 3) set EMBEDDINGS_URL to a trusted localhost endpoint and verify the script enforces localhost-only. Also consider adding 'curl' to required-binaries or removing the curl usage and re-running an audit. If you cannot audit or sandbox the skill, treat it as potentially able to exfiltrate data and avoid giving it access to sensitive transcripts.
能力评估
Purpose & Capability
The skill's name/description (a metacognition engine) matches the provided code: it stores categorized insights, applies decay, builds a graph, and compiles a lens. However, the code depends on an external 'curl' binary (via subprocess) to call an embeddings endpoint even though required binaries list only python3; requiring curl was not declared.
Instruction Scope
SKILL.md and README explicitly state 'no curl/subprocess' and 'local-only embeddings validated at startup', but metacognition.py uses subprocess.run to call curl for the embeddings endpoint and does not contain code that validates EMBEDDINGS_URL to localhost-only. This is a direct contradiction: runtime instructions promise no subprocess and local-only network, but the script performs network calls via curl and will honour whatever EMBEDDINGS_URL is set to (including remote URLs) unless the environment is constrained.
Install Mechanism
There is no install spec (instruction-only plus a code file), which keeps disk/write risk low. Still, the script executes an external binary (curl) if present — the skill package did not declare curl as a required binary, so the runtime will silently rely on an unlisted dependency. No archive downloads or remote installers are present.
Credentials
The registry lists no required env vars, but the script reads EMBEDDINGS_URL from the environment and will attempt network calls to it. SKILL.md/README claim embedded endpoint is optional and validated to localhost, but the code does not implement that validation. If EMBEDDINGS_URL is set to a remote server, the script will send text (potentially session content) to that endpoint — an environment variable can thus enable exfiltration. The skill does write to declared local paths (memory/ and scripts/).
Persistence & Privilege
always is false and the skill is user-invocable; it writes only to workspace-relative paths (memory/metacognition.json and scripts/metacognition-lens.md) advertised in SKILL.md. It does not request system-wide privileges or modify other skills' configs.
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 nate-metacognition - 安装完成后，直接呼叫该 Skill 的名称或使用
/nate-metacognition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
v1.0.0
Initial release of the Metacognition skill — a self-reflection engine for AI agents.
- Extracts and reinforces patterns from session transcripts using Hebbian learning and time decay.
- Maintains a categorized knowledge store, building a weighted graph of insights and connections.
- Compiles a token-limited "metacognition lens" summarizing active self-knowledge.
- Provides CLI commands for maintaining, updating, and summarizing insights.
- Supports optional local embeddings for semantic similarity; designed for secure, local use.
元数据
常见问题
Metacognition 是什么？
Self-reflection engine for AI agents. Extracts patterns from session transcripts into a weighted graph with Hebbian learning and time decay. Compiles a token... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 89 次。
如何安装 Metacognition？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install nate-metacognition」即可一键安装，无需额外配置。
Metacognition 是免费的吗？
是的，Metacognition 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Metacognition 支持哪些平台？
Metacognition 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 Metacognition？
由 JoeyCacciatore3（@joeycacciatore3）开发并维护，当前版本 v1.0.0。
推荐 Skills