← 返回 Skills 市场

Karpathy Query Feedback

Name: Karpathy Query Feedback
Author: sora-mury

作者 sune · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ✓ 安全检测通过

总下载

当前安装

版本数

在 OpenClaw 中安装

/install karpathy-query-feedback

功能描述

Execute Karpathy LLM queries by searching memories via M-Flow, formatting results as wiki entries, and saving them for later compilation.

使用说明 (SKILL.md)

Karpathy Query → Wiki 回流 Skill

描述

实现 Karpathy LLM Knowledge Base 的第一阶段：Query → Wiki回流。

当用户发起查询时：

使用 M-Flow 搜索相关记忆
将结果格式化为 wiki 条目
存入 wiki 层供后续 Compile 使用

激活条件

用户发起知识查询
需要将查询结果回流到 wiki
session → knowledge pipeline

工作流

用户查询 → M-Flow搜索 → Wiki格式化 → 存入wiki层 → 供Compile使用

Wiki 条目格式

| source | content | tags | timestamp |
|--------|---------|------|-----------|
| session:xxx | 知识内容 | tag1,tag2 | 2026-04-05 |

使用方式

Python API

from karpathy_query_feedback import QueryFeedbackPipeline

pipeline = QueryFeedbackPipeline()
results = await pipeline.query("用户询问的问题")
wiki_entries = pipeline.format_as_wiki(results)
await pipeline.save_to_wiki(wiki_entries)

命令行

python scripts/query_and_save.py "查询内容" --format wiki --output ./wiki/

搜索模式

lexical: BM25 全文搜索（快速、精确）
episodic: 向量搜索（语义相似）
triplet: 三元组搜索（关系推理）
hybrid: 混合搜索（lexical + episodic）

配置

使用 M-Flow 作为底层记忆系统
Wiki 存储路径: knowledge/wiki/
标签体系: 从配置或自动提取

依赖

m-flow-memory skill (已安装)
knowledge-distillation skill (用于标签提取)

文件结构

karpathy-query-feedback/
├── SKILL.md
├── scripts/
│   ├── __init__.py
│   ├── pipeline.py      # 核心管道
│   ├── formatter.py     # Wiki格式化
│   ├── search.py        # 搜索封装
│   └── query_and_save.py # CLI入口
└── docs/
    └── README.md

安全使用建议

This skill appears coherent and limited to local memory search and file writes. Before installing: 1) Verify and trust the declared dependency 'm-flow-memory' (the skill dynamically loads and executes that module from disk). 2) Be aware it will create/append files under knowledge/wiki/ in your environment. 3) There are no requested secrets or network calls in the provided code, but if you use this in a shared or production environment, review the m-flow and knowledge-distillation skills for any network or credential usage. If unsure, run it in a sandbox or inspect the m-flow dependency first.

功能分析

Type: OpenClaw Skill Name: karpathy-query-feedback Version: 1.0.0 The skill bundle implements a knowledge management pipeline that retrieves information from a memory system (M-Flow) and saves it as formatted Markdown wiki entries. The code in scripts/__init__.py and the various test scripts (test_pipeline.py, debug_search.py) perform standard file I/O and inter-skill communication within the expected OpenClaw environment, with no evidence of data exfiltration, malicious execution, or prompt injection.

能力评估

✓ Purpose & Capability

Name/description (Karpathy Query → Wiki回流) match the implementation: the pipeline searches an M‑Flow memory, converts results to WikiEntry objects, and saves them under knowledge/wiki/. Declared dependency on m-flow-memory is appropriate for this functionality.

✓ Instruction Scope

SKILL.md and scripts consistently instruct the agent to use M‑Flow, format entries as markdown table rows, and save them to a local wiki path. The code only reads/writes files under the skill's repo (knowledge/wiki/) and imports the local m-flow skill; there are no instructions to read unrelated system files, environment variables, or to transmit data to external endpoints.

✓ Install Mechanism

No install spec is provided (instruction/code bundle only). There are no downloads or external installers; code is included in the skill. This is the lowest-risk install pattern for this platform.

✓ Credentials

The skill requires no environment variables, credentials, or special config paths. It does dynamically import the m-flow skill from a relative path (declared as a dependency), which is expected and proportional to its purpose.

ℹ Persistence & Privilege

The skill creates and appends markdown files under knowledge/wiki/ (self-persistence of produced wiki content). always:false (not force-installed) and it does not modify other skills' configurations, but it will execute code from the local m-flow dependency at runtime via importlib (this is normal for code dependencies—review that dependency before trusting).

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 karpathy-query-feedback
安装完成后，直接呼叫该 Skill 的名称或使用 /karpathy-query-feedback 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: Query to Wiki feedback loop

元数据

Slug karpathy-query-feedback

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题