Cortex — Graph Memory Skill

You have access to Cortex, a self-organizing knowledge graph for persistent memory. Use it to remember facts, decisions, goals, patterns, and observations across sessions. Knowledge is stored as nodes in a graph that auto-links, decays stale information, detects contradictions, and computes trust from topology.

When to Use Cortex

• Start of session: Call cortex_briefing to load context from previous sessions.
• Learning something important: Call cortex_store to persist facts, decisions, goals, events, patterns, or observations.
• Answering questions about past work: Call cortex_search or cortex_recall to find relevant knowledge.
• Understanding relationships: Call cortex_traverse to explore how concepts connect.
• Connecting ideas: Call cortex_relate to explicitly link related nodes.

Tools Reference

cortex_store — Remember something

Store a knowledge node. Cortex auto-generates embeddings and the auto-linker discovers connections in the background.

cortex_store(
  title: string,         # Required. Short summary (used for search and dedup).
  kind: string,          # "fact" | "decision" | "goal" | "event" | "pattern" | "observation" | "preference". Default: "fact"
  body: string,          # Full content. Can be long. Include details here.
  tags: string[],        # Optional tags for filtering.
  importance: number     # 0.0–1.0. Higher = retained longer, weighted more. Default: 0.5
)

Returns: { id, message }.

Guidelines:

• Use importance >= 0.7 for architectural decisions, credentials, project goals, user preferences.
• Use importance 0.4–0.6 for routine facts, observations, intermediate findings.
• Use importance <= 0.3 for ephemeral notes, temporary context.
• Write titles as self-contained statements: "API uses JWT authentication" not "Auth info".
• Put details, reasoning, and evidence in body.
• Use accurate kind values — they affect briefing structure and filtering.
• Tag with project name, domain, or agent role for scoped retrieval.

cortex_search — Find by meaning

Semantic similarity search across all stored knowledge.

cortex_search(
  query: string,   # Required. Natural language query.
  limit: integer,  # Max results. Default: 10
  kind: string     # Optional filter: "fact", "decision", "goal", etc.
)

Returns: array of { id, kind, title, body, score, created_at }.

When to use: Quick lookup of specific facts or concepts. Best when you know roughly what you're looking for.

cortex_recall — Contextual retrieval

Hybrid search combining vector similarity AND graph structure. Returns more contextually relevant results than pure search.

cortex_recall(
  query: string,   # Required. What to recall.
  limit: integer,  # Default: 10
  alpha: number    # 0.0 = pure graph, 1.0 = pure vector.