Local-first 🧠memory for Hermes Agent that lives in your Obsidian vault and remembers project context. Nightly 😴 dream passes turn repeat corrections into confirmed preferences with measurable confidence. Adapters ship for Claude Code, Codex, and OpenClaw, with an MCP server for anything else.
CodeGraph and MCP operational readability implementation plan
Branch: feat/codegraph-mcp-operational-readability
Design: docs/brainstorm/codegraph-mcp-operational-readability/design.md
t_a1e76788 - [upstream:graphify] Cargo workspace dependency extraction for Rust projects
Files
src/core/partner/codegraph.tsorsrc/core/partner/codegraph-report.tssrc/cli/...for theo2b partner codegraph reportcommand and help textsrc/mcp/brain/knowledge-tools.tsor the current MCP registration module forbrain_codegraph_reporttests/core/partner/codegraph.test.ts- Focused CLI or MCP tests for the report wrapper
docs/cli-reference.mddocs/mcp.mdCHANGELOG.md
Acceptance
A focused test suite passes that proves:
- A Rust project with a structural
Cargo.tomlworkspace reports workspace members. - A non-Rust project reports
cargo_workspace: nullwith an explicit reason. - Missing
codegraphCLI and missing.codegraph/are represented as honest report states rather than thrown failures. - The CLI and MCP wrappers return the core report shape without mutating the project or vault.
Suggested verification command for the worker:
bun test tests/core/partner/codegraph.test.ts <new focused CLI or MCP test file>
bun run typecheck
Depends on
No sibling task needs to land first. Build this as the first card if possible, because it establishes the dedicated report surface and naming conventions that the release docs can reuse.
t_a286135c - [upstream:graphify] Multi-batch community labeling for large graphs
Files
src/core/brain/link-graph/communities.tssrc/cli/brain/verbs/clusters.tssrc/mcp/brain/knowledge-tools.tstests/cli/brain-clusters.test.tstests/mcp/link-recall-tools.test.tsor the current MCP cluster test filedocs/cli-reference.mddocs/mcp.mdCHANGELOG.md
Acceptance
A focused test suite passes that proves:
- Existing
brain_clustersando2b brain clusters runoutput remains unchanged when no batch option is supplied. - Invalid batch sizes fail with typed validation or usage errors.
- Batched mode processes cluster note materialization in bounded chunks and reports per-batch successes and failures explicitly.
- The implementation does not introduce LLM labeling, natural-language classification, or a Graphify dependency.
Suggested verification command for the worker:
bun test tests/cli/brain-clusters.test.ts <new focused MCP test if added>
bun run typecheck
Depends on
Prefer running after t_a1e76788 on the same branch so shared release documentation and changelog entries can build on the CodeGraph report wording. The implementation itself should not depend on the CodeGraph report code.