version: "1.0.0" name: ciel-knowledge-and-memory description: CIEL's framework for multi-layer knowledge ingestion, semantic memory search, and structural code exploration. license: MIT metadata: ciel-version: 1.0.0 ciel-extension: ciel.yaml

CIEL ADAPTATION: Knowledge & Memory (The Intel Layer)

This skill manages the flow of information from live codebase structure to durable cross-session memory.

Knowledge Ingestion (Layers)

1. Active (L1): GitHub/Linear for live roadmap and implementation truth.
2. Short-term (L2): Project memory files (~/.claude/projects/*/memory/) for user preferences.
3. Semantic (L3): MCP Memory Server for structured knowledge graphs and observations.
4. Durable (L4): Dedicated knowledge base repo for curated research and synthesized docs.

Memory Search (Index -> Timeline -> Fetch)

• Mandate: NEVER fetch full details without filtering first.
• Protocol:

1. Search: Get a table of IDs and Titles to minimize tokens.
2. Timeline: Interleave sessions and observations around an anchor ID to find context.
3. Fetch: Batch-get only the 2-3 most relevant full observations.

Structural Exploration (Smart Explore)

• Map First: Use smart_search or smart_outline (AST-based) before performing a full read_file.
• Unfold: Review signatures and unfold ONLY the implementation of specific functions/methods needed.
• Token Efficiency: Target 4-8x token savings vs raw file reads.

Anti-Patterns

• The Context Dump: Reading 10+ full files when a structural map would suffice.
• Amnesiac Session: Starting work without checking MCP memory for previous decisions.
• Durable-Code Drift: Storing live application code in a knowledge base instead of the Git repo.