Phase 3: Brain Digest + Brain Entity — Implementation Plan
For Claude: REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
Goal: Add brain_digest MCP tool (structured content ingestion with entity/relation/action extraction) and brain_entity MCP tool (entity lookup with evidence).
Architecture: brain_digest creates a new chunk from input content, runs Phase 2's entity extraction pipeline on it, extracts action items/decisions/questions via LLM, applies Phase 6's sentiment analysis, stores everything in KG tables with confidence tiers and user_verified flags. brain_entity is a read-only lookup that returns structured entity info with relations and evidence chunks.
Tech Stack: Python, APSW/sqlite-vec, Phase 2 extraction pipeline, Phase 6 sentiment, Ollama/MLX LLM
Task 1: Add user_verified column to KG tables
Files:
- Modify: src/brainlayer/vector_store.py (add columns to kg_entities and kg_relations)
- Test: tests/test_phase3_digest.py (new file)
Task 2: Digest pipeline module — extract structured knowledge from text
Files:
- Create: src/brainlayer/pipeline/digest.py
- Test: tests/test_phase3_digest.py
Core function: digest_content(content, store, embed_fn, ...) that:
1. Creates a chunk with source="digest"
2. Runs entity extraction (Phase 2's process_chunk + store_extraction_result)
3. Runs sentiment analysis (Phase 6's analyze_sentiment)
4. Extracts action_items, decisions, questions via LLM
5. Applies confidence tiers
6. Returns structured DigestResult
Task 3: brain_digest MCP tool
Files:
- Modify: src/brainlayer/mcp/__init__.py (add brain_digest tool + handler)
- Test: tests/test_phase3_digest.py
Task 4: brain_entity MCP tool
Files:
- Modify: src/brainlayer/mcp/__init__.py (add brain_entity tool + handler)
- Test: tests/test_phase3_digest.py
Task 5: CLI command + integration test
Files:
- Modify: src/brainlayer/cli/__init__.py (add digest command)
- Test: tests/test_phase3_digest.py