AgentSkillsCN

recall

在项目记忆库中搜索相关上下文。可用于查找过往决策、相似问题或会话历史。触发条件包括：回忆、回想、我们当时做了什么决定、在记忆中查找、检索记忆、回顾过往会话。

SKILL.md
--- frontmatter
name: recall
description: "Search project memory for relevant context. Use to find past decisions, similar problems, or session history. Triggers on: recall, remember, what did we decide, find in memory, search memory, past sessions."

Memory Recall

Search project memory to find relevant past context.

The Job

  1. •Understand what the user is looking for
  2. •Search across memory sources
  3. •Return relevant context with references

Memory Sources

Search in order of relevance:

1. Project Memory (~/aiconfig/memory/projects/{project}/)

  • •context.json - Current project state, tech stack, known issues
  • •sessions.json - Past session summaries and follow-ups
  • •decisions.json - Architectural Decision Records

2. Global Memory (~/aiconfig/memory/global/)

  • •preferences.json - Cross-project coding preferences
  • •patterns.json - Learned patterns and solutions

3. Vector Search (if available)

  • •LanceDB embeddings for semantic search
  • •"Find similar problems I've solved"

Search Strategies

By Topic

code
User: "What did we decide about authentication?"

Search:
1. decisions.json for title/context containing "auth"
2. sessions.json for decisions_made mentioning "auth"
3. context.json for relevant architecture notes

By Time

code
User: "What did we do last session?"

Search:
1. sessions.json, sort by date, return most recent

By File

code
User: "What changes have we made to the auth module?"

Search:
1. sessions.json for files_modified matching pattern
2. Return sessions that touched those files

Semantic (with LanceDB)

code
User: "Have I solved a similar caching problem before?"

Search:
1. Embed the query
2. Vector search across session summaries and decisions
3. Return top matches with similarity scores

Output Format

For Decisions

code
Found 2 relevant decisions:

**ADR-003: Use Redis for session storage** (2025-01-10)
- Context: Need fast session lookups for auth
- Decision: Redis with 24h TTL
- Rationale: Sub-millisecond reads, built-in expiry

**ADR-007: JWT for API authentication** (2025-01-15)
- Context: API needs stateless auth
- Decision: Short-lived JWTs with refresh tokens
- Rationale: Scalable, no session storage needed

For Sessions

code
Found 3 sessions mentioning "auth":

**2025-01-15** - Implemented JWT authentication
- Files: src/auth/jwt.ts, src/middleware/auth.ts
- Follow-up: Add refresh token rotation

**2025-01-12** - Set up user model
- Files: prisma/schema.prisma, src/models/user.ts
- Decisions: Used bcrypt for password hashing

For Context

code
Current project context:

**Tech Stack**: Next.js, TypeScript, Prisma, PostgreSQL
**Current Focus**: Payment integration
**Known Issues**:
- Slow initial page load (investigating)
- Auth tokens not refreshing properly

No Results

If nothing found:

code
No memory found for "{query}".

This could mean:
1. We haven't worked on this topic yet
2. It wasn't logged in a previous session
3. Try different search terms

Would you like to:
A) Search with different terms
B) Check global preferences
C) Start fresh (I'll help figure it out)

Checklist

  • • Understood what user is looking for
  • • Searched relevant memory sources
  • • Returned context with clear references
  • • Offered alternatives if nothing found