1. Delete dead cleanup_deleted_files() (zero callers)
2. Wire hooks/diff.rs into edit extractor — compute_diff() now called
   with old/new strings, semantic summary added to memory content+metadata
3. Replace panic!() in rule_loader.rs with Result return
4. Add "summarize" to CLI VALID_CYCLES (was unreachable from CLI)
5. Fix MCP consolidate auto mode to run all 5 cycles
6. Delete unused GraphConfig fields (max_expansion_hops, bridge_similarity_threshold)
7. Delete unused StorageConfig.db_path field
8. Fix stale "30 tools" comment → 32 in mcp/mod.rs and test assertions
9. Narrow test-only pub functions with #[cfg(test)]: personalized_pagerank,
   strongly_connected_components, max_degree, graph_memory_estimate,
   Bm25Index::build, OpenAIProvider::with_api_key
10. Extract MemoryRow::from_row() — eliminates 4 copy-paste blocks

Also: fix comparison.md recommendations (claude-mem, Cognee, AutoMem,
Supermemory, claude-context, MCP tool count 28→32)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

…gonomics (#14)

- Add MemoryNode::new() and test_default() constructors, replacing 63 struct literals
- Add MapStorageErr trait, replacing 182 .map_err(|e| CodememError::Storage(...)) calls
- Move truncate to codemem-core::utils, re-export from scoring.rs and codemem lib
- Extract insert_memory_inner() helper, deduplicating insert_memory/insert_memory_no_tx
- Merge store_insight/store_pattern_memory into shared store_enrichment_memory helper
- Convert ~20 test files to use MemoryNode::test_default()

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

…ings deps (#15)

* refactor: tier 3 — move domain logic to engine, eliminate binary's direct storage/embeddings deps

Tasks 23-28 plus the dependency elimination:

- store_memory_with_links: consolidated edge building + auto-linking into engine
- Insight aggregation: moved PageRank/Louvain/topology from API routes to engine
- Eliminated storage_direct from API: all access routed through engine
- Moved PCA to engine crate (computational logic, not transport)
- Added engine facade methods for graph ops (traverse, stats, edges, louvain, pagerank)
- Moved RawGraphMetrics to codemem-core as cross-crate DTO
- Removed codemem-storage and codemem-embeddings from binary crate deps;
  all access now via re-exports from codemem-engine
- Simplified build_server() from ~50 lines of manual assembly to
  McpServer::from_db_path() delegation
- Made CodememEngine::from_db_path() handle corrupt vector files gracefully

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: resolve 6 infinite recursion bugs in StorageBackend trait impl

Use fully-qualified `Storage::method(self, ...)` syntax to dispatch to
the inherent impl instead of recursing into the trait method.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: remove redundant save_index() in store_memory_with_links

persist_memory() already calls save_index() internally via
persist_memory_inner(memory, true), so the extra call serialized
the vector index to disk twice on every store_memory MCP call.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

…#16)

* refactor: fix bugs, unify flows, add safe accessors, define constants

Phase 1 — Bug fixes:
- Fix search_code double-inverting HNSW similarity scores
- Fix delete_namespace using non-atomic deletes (now cascade)
- Fix session_checkpoint bypassing BM25/graph/embedding pipeline
- Fix CLI export bumping access_count on read-only operation
- Fix API PUT /memories double access_count bump

Phase 2 — Safe read-only access:
- Add with_graph()/with_vector() closure accessors on CodememEngine
- Remove raw mutex accessors (graph_mutex, vector_mutex, etc.)

Phase 3 — Unify duplicate flows:
- MCP session_context delegates to engine.session_context()
- API namespace stats delegates to engine.namespace_stats()
- Unified analyze pipeline via engine.analyze(AnalyzeOptions)
- Extract collect_subgraph() helper for BFS+edge pattern in API
- All API graph routes use engine.with_graph() instead of raw mutex
- Remove MCP delegation methods (lock_graph, lock_vector, etc.)

Phase 4 — Audit cleanup:
- Define named constants for scoring weights in scoring.rs
- Define named constants for compaction weights in compaction.rs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* style: address Aikido review — reduce nesting, extract helpers, early return

- Extract collect_edges_between() helper to flatten collect_subgraph nesting
- Extract build_subgraph_response() to move logic out of with_graph closure
- Use early return in tool_session_context instead of match arm nesting

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: introduce RecallQuery struct, decompose persist pipeline

Item 35: Replace recall()'s 7 positional parameters with RecallQuery struct.
Named fields prevent argument swaps; RecallQuery::new() provides a clean
default for the common case. Extract passes_quality_filters() helper.
Update all 20+ callers across engine, MCP, API, and tests.

Item 31: Break persist_index_results_with_progress (490 lines) into focused
helpers: persist_graph_nodes, build_package_tree, build_symbol_nodes,
build_symbol_payload, build_reference_edges, build_chunk_nodes,
embed_and_persist, persist_nodes/edges_to_storage_and_graph.
Orchestrator is now ~30 lines.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* style: flatten build_package_tree nesting with early-continue guards

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>