Memory Recall Skill

Search and surface relevant memories to inform current work with decisions, patterns, and learnings from past sessions.

Trigger Phrases

• "what did we decide about", "how do we handle"
• "find memories about", "search for", "recall"
• "what's our approach to", "what patterns do we use"
• "any gotchas with", "lessons learned about"
• "previous decisions on", "remind me"

Quick Reference

Search Mode	Best For	Example
hybrid (default)	General queries, balanced results	"database storage decision"
vector	Conceptual similarity, fuzzy matching	"how to handle errors gracefully"
text	Exact terms, specific keywords	"PostgreSQL"

Execution Strategy

<strategy> **MCP-First Approach:** 1. Use `mcp__subcog__subcog_recall` tool when available 2. Fall back to CLI: `subcog recall "<query>" --mode <mode> --limit <n>`

Search Quality Guidelines:

• Start broad, then narrow with namespace filters
• Use vector mode for conceptual searches
• Use text mode when you know exact terms
• hybrid mode (default) works best for most queries </strategy>

Intelligent Search Workflow

<workflow> When searching memories:

1. Understand the intent - what does the user need to know?
2. Identify keywords - extract searchable terms
3. Choose search mode based on query type
4. Apply namespace filter if domain is clear
5. Execute search via MCP tool or CLI
6. Interpret results - explain relevance scores
7. Synthesize findings - summarize key insights </workflow>

Score Interpretation

<scores> | Score Range | Meaning | Action | |-------------|---------|--------| | 0.9 - 1.0 | Exact or near-exact match | High confidence | | 0.7 - 0.9 | Strong relevance | Good match | | 0.5 - 0.7 | Moderate relevance | Related but may need refinement | | < 0.5 | Low relevance | Try different query | </scores>

Search Strategies by Use Case

<strategies> **Finding a specific decision:** ``` Mode: text or hybrid Namespace: decisions Example: "PostgreSQL storage decision" ```

Finding patterns for a domain:

Mode: vector (conceptual similarity)
Namespace: patterns
Example: "how to handle API errors gracefully"

Debugging help (gotchas):

Mode: hybrid
Namespace: learnings
Example: "authentication token refresh issues"

</strategies>

MCP Tool Reference

<mcp> **Tool:** `mcp__subcog__subcog_recall`

Parameters:

• query (required): Natural language search query
• mode (optional): "hybrid" (default), "vector", or "text"
• namespace (optional): Filter to specific namespace
• limit (optional): Maximum results (default: 10, max: 50)
• format (optional): Output format for MIF (Memory Interchange Format)
  • json (default): Single-line JSON-LD for token efficiency
  • detail: Pretty-printed JSON-LD for human readability
  • md: Markdown with YAML frontmatter
  • compact: Single-line JSON for streaming

Returns:

• Array of memories in the specified MIF format, each including:
  • @id: URN identifier (e.g., urn:mif:memory-id)
  • @type: "Memory"
  • content, namespace, tags, created, modified
  • subcog:score: Relevance score (0.0-1.0) </mcp>