Pre-release doc-staleness sweep for 2.5.0:
- ARCHITECTURE.md config reference: add the [maintenance] section
(enabled/forget_sweep/lint/embedding_backfill intervals +
reconcile_tombstones_deleted_pages, #929/#964) and release_base_url (#801).
- README support matrix: Hermes Agent Community -> Supported, aligning the
compact table with the authoritative docs/support-matrix.md (#933 hooks +
#942 skills are first-party).
- research-codebase-memory-mcp.md: correct the stale '19 tools' -> 23.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MDbhmszrjG9s5MrPrTuNtm
Ground the three borrowable ideas against the current tree:
- tiered tool profiles -> 2.1 feature (list_tools single choke point,
read/write partition already exists via tool_call_is_write)
- token-efficiency metric -> land on main any time (additive, eval-only)
- committed compressed index -> defer (server-less distribution model,
needs a concrete use-case first)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MDbhmszrjG9s5MrPrTuNtm
codebase-memory-mcp (github.com/DeusData/codebase-memory-mcp, ~42k
stars, pure-C static binary) is a code-intelligence MCP: tree-sitter
over 162 languages → a SQLite knowledge graph queried with openCypher,
~120x fewer tokens than grep. It remembers the *code's structure*, not
the *session* — complementary to ai-memory, not a competitor (an agent
can run both). New docs/research-codebase-memory-mcp.md; transferable
ideas noted: tiered tool profiles (Scout/Verify/Auditor) to cut MCP
over-permissioning, a published token-efficiency metric, and an optional
git-committed compressed index for server-less teams. Linked from the
landscape and ARCHITECTURE prior-art list.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MDbhmszrjG9s5MrPrTuNtm
