code-recall

This skill helps you interact with the code-recall MCP server for persistent memory across coding sessions.

Session Start

At the beginning of each coding session, call get_briefing to:

• Get statistics about stored memories
• See recent warnings and failed approaches
• Pre-fetch context for your focus areas

{
  "tool": "get_briefing",
  "arguments": {
    "focus_areas": ["authentication", "database"]
  }
}

Before Making Changes

Before implementing significant changes:

1. Search past decisions with search_memory:

{
  "tool": "search_memory",
  "arguments": {
    "query": "authentication approach",
    "category": "decision"
  }
}

2. Check applicable rules with check_rules:

{
  "tool": "check_rules",
  "arguments": {
    "action": "adding API endpoint for user authentication"
  }
}

After Making Decisions

After making architectural or important decisions, store them with store_observation:

{
  "tool": "store_observation",
  "arguments": {
    "category": "decision",
    "content": "Use JWT for authentication instead of sessions",
    "rationale": "Stateless auth enables horizontal scaling and works better with our microservices architecture",
    "tags": ["auth", "jwt", "architecture"]
  }
}

Categories

• decision - Architectural or implementation choices
• pattern - Reusable patterns that work well
• warning - Things to avoid or be careful about
• learning - Insights gained from implementation

After Implementation

Once you verify if something worked or failed, record the outcome:

{
  "tool": "record_outcome",
  "arguments": {
    "memory_id": 1,
    "outcome": "JWT implementation works well, token refresh handled smoothly",
    "worked": true
  }
}

Important: Setting worked: false boosts that memory in future searches, helping avoid repeated mistakes.

Creating Rules

To establish project guardrails, use set_rule:

{
  "tool": "set_rule",
  "arguments": {
    "trigger": "adding API endpoint",
    "must_do": ["Add rate limiting", "Write integration tests", "Update API documentation"],
    "must_not": ["Skip authentication middleware"],
    "ask_first": ["Is this a breaking change?", "Does this need versioning?"]
  }
}

Example Workflow

1. Session starts

   get_briefing({ focus_areas: ["payments"] })
   

2. About to implement payments

   search_memory({ query: "payment integration" })
   check_rules({ action: "adding payment processing" })
   

3. Decide on Stripe

   store_observation({
     category: "decision",
     content: "Use Stripe for payments",
     rationale: "Best documentation, webhook support, and PCI compliance handled"
   })
   

4. After implementation works

   record_outcome({
     memory_id: 5,
     outcome: "Stripe integration complete, webhooks working reliably",
     worked: true
   })
   

When to Store Memories

Do store:

• Architectural decisions and their rationale
• Patterns that proved useful
• Approaches that failed (with worked: false)
• Project-specific rules and conventions

Don't store:

• Trivial implementation details
• Temporary debugging notes
• Information already in documentation