← Back to Skills Marketplace
109
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install karpathy-lint
Description
Performs quality checks on Karpathy LLM knowledge points including deduplication, merging, updating, and generates reports on knowledge base health.
README (SKILL.md)
Karpathy Lint Skill - Phase 3
描述
实现 Karpathy LLM Knowledge Base 的第三阶段：知识自检与修复。
对 Phase 2 生成的 knowledge points 进行质量检查：去重、合并、更新、删除。
工作流程
Phase 1: 用户查询 → Wiki 条目
Phase 2: Wiki → Knowledge Points (精炼)
Phase 3: Lint → 去重/合并/更新/自修复 ← 当前
Phase 4: 高级 (M-Flow 集成、多Agent共享)
Lint 检查项
-
去重 (Deduplication)
- 检测相似/重复的 knowledge points
- 合并高度相似的内容
-
更新 (Update)
- 检查知识点是否过期
- 标记需要更新的条目
-
质量检查 (Quality Check)
- 检查内容完整性
- 验证标签一致性
-
生成健康报告
- 当前知识库统计
- 发现的问题列表
- 修复建议
文件结构
karpathy-lint/
├── SKILL.md
├── scripts/
│ ├── __init__.py # LintPipeline
│ ├── dedup.py # 去重逻辑
│ ├── merger.py # 合并逻辑
│ └── test_lint.py # 测试
依赖
- Phase 2 的 knowledge points 文件
- M-Flow (可选，用于存储)
Usage Guidance
Before installing or running: (1) Confirm the repository includes the missing modules (dedup.py, merger.py or equivalent) — the SKILL.md lists files that are not present. (2) Inspect the remainder of scripts/__init__.py (it was truncated in the review) for any network calls, secrets exfiltration, or code that calls external services (M-Flow integration was mentioned). (3) Note the default behavior: it will read from and write to a sibling/parent 'knowledge/knowledge-points' directory (outside the skill folder). If that path is not where you want it to operate, run it with an explicit kp_path pointing to a safe sandbox copy. (4) Run the tests in a sandboxed environment or container with only sample knowledge files to observe behavior. (5) If you don't trust the unknown source, request a homepage or source repo and a complete file listing from the publisher before use.
Capability Analysis
Type: OpenClaw Skill
Name: karpathy-lint
Version: 1.0.0
The skill bundle implements a knowledge base linting and deduplication pipeline for Markdown files. The code in scripts/__init__.py and scripts/test_lint.py performs legitimate text processing, similarity calculations, and report generation without any signs of data exfiltration, malicious execution, or prompt injection.
Capability Assessment
Purpose & Capability
The SKILL.md describes deduplication, merging, and merging helper modules (dedup.py, merger.py), but the package only includes scripts/__init__.py and test_lint.py — the named helper modules are missing. That mismatch suggests the bundle is incomplete or the implementation is different from the documentation.
Instruction Scope
The code's KnowledgePointParser defaults to a path three levels up plus /knowledge/knowledge-points, i.e. it reads and writes files outside the skill folder (repository-root/knowledge/knowledge-points). This will access user filesystem data outside the skill directory and write lint-report.md there. The SKILL.md mentions Phase 2 knowledge files but does not make the exact path/side-effects explicit.
Install Mechanism
No install steps or external downloads are present (instruction-only with included scripts), so there is no installer that would fetch remote code. This reduces installation-time supply-chain risk.
Credentials
The skill doesn't request any environment variables or external credentials, which is proportional to its stated purpose. However, because it directly reads and writes files outside its own directory, it effectively requires filesystem access to the user's knowledge directory — confirm that the default path matches where you want it to operate or that you will pass an explicit, safe path.
Persistence & Privilege
The skill is not always-enabled and does not declare system-wide privileges. It does write reports and can save updated knowledge-point files into the inferred knowledge directory, which is expected for a lint/repair tool but is a form of persistent modification to user data (not to other skills or system configs).
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install karpathy-lint - After installation, invoke the skill by name or use
/karpathy-lint - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: Weekly memory health check and maintenance
Metadata
Frequently Asked Questions
What is Karpathy Lint?
Performs quality checks on Karpathy LLM knowledge points including deduplication, merging, updating, and generates reports on knowledge base health. It is an AI Agent Skill for Claude Code / OpenClaw, with 109 downloads so far.
How do I install Karpathy Lint?
Run "/install karpathy-lint" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Karpathy Lint free?
Yes, Karpathy Lint is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Karpathy Lint support?
Karpathy Lint is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Karpathy Lint?
It is built and maintained by sune (@sora-mury); the current version is v1.0.0.
More Skills