liuzhengdongfortest/cs-learn

What does cs-learn do?

Turn pitfalls or best practices into a searchable learning archive, with two tracks: pitfall (mistakes) and knowledge (default practices). Triggered when the user says "沉淀知识", "learning", "把这次经验记下来", or when an acceptance/fix is finalized.

How It Works

The skill listens for specific trigger phrases in the conversation:

• "沉淀知识" – Chinese for "accumulate knowledge"
• "learning"
• "把这次经验记下来" – Chinese for "record this experience"
• Acceptance criteria or fix completion events

Once triggered, the AI assistant will ask the user a series of questions to structure the learning document:

1. Title – A short, descriptive title for the learning entry.
2. Track – Choose either pitfall (for mistakes and lessons learned) or knowledge (for best practices and default approaches).
3. Summary – A one-sentence summary of what was learned.
4. Detailed Description – A more thorough explanation of the context, the problem, the solution, and any important details.
5. Tags – Optional keywords to make the entry easily searchable later.

After the user provides the information, the skill saves the document as a Markdown file in the appropriate directory under docs/learning/:

• docs/learning/pitfall/<slug>.md – for pitfalls
• docs/learning/knowledge/<slug>.md – for knowledge

The file includes YAML frontmatter with the title, track, tags, and a timestamp.

Example

User: "沉淀知识: When using GraphQL, avoid N+1 queries by using DataLoader"

AI: Prompts for title, track, summary, details, and tags, then creates: docs/learning/knowledge/avoid-n-plus-1-graphql.md

---
title: Avoid N+1 Queries in GraphQL with DataLoader
track: knowledge
date: 2025-03-29
tags:
  - graphql
  - dataloader
  - n-plus-1
  - performance
---

# Avoid N+1 Queries in GraphQL with DataLoader

**Summary:** Always use DataLoader to batch and cache database queries in GraphQL resolvers to prevent the N+1 query problem.

**Details:**
...

Benefits

• Searchable – Easily find past learnings using tags or full-text search.
• Structured – Clear separation between mistakes and best practices.
• Team-friendly – Shareable Markdown files that can be committed to version control.
• Contextual – Captures the "why" behind decisions, not just the code.

Supported Platforms

Designed for AI agents like Claude Code, Cursor, and Codex.