Engineering Memory recipes (MCP)¶

Ranked scope context and governed drafts — not a second analyzer. Normative tool shapes: Engineering Memory MCP surface.

Session-local review markers live in Coverage join & session markers.

1. Bootstrap before first scoped retrieval¶

When the store is missing, default mcp_sync_policy=bootstrap_if_missing ingests from the latest MCP run on the first scoped get_relevant_memory.

Step	Tool / action
Analyze	`analyze_repository(root=<abs>)`
Optional explicit ingest	`manage_engineering_memory(action=refresh_from_run, root=<abs>)`
Offline init	`codeclone memory init` (CI/offline; same ingest contract)

2. Scope context after `start_controlled_change`¶

Call only after edit_allowed=true. root is required (same absolute path as analysis).

get_relevant_memory(root=<abs>, intent_id=<id>)
  # or scope=["path/to/file.py", ...]

Read memory_sync, stale warnings, and contradiction_note entries before editing. Do not treat draft / inferred rows as established facts.

3. Draft observations during the cycle¶

manage_engineering_memory(
  action=record_candidate,
  root=<abs>,
  record_type=risk_note | change_rationale | ...,
  statement="<one durable fact>",
  subject_path="path/to/main/file.py",
)

Agents cannot approve / reject / archive via MCP. Humans promote drafts in the VS Code Memory view or with codeclone memory approve --i-know-what-im-doing (optional --by NAME).

4. Finish proposals¶

On accepted finish:

finish_controlled_change(..., propose_memory=true)

Returns memory_candidates, memory_staleness, memory_coverage_delta, and may enqueue projection rebuild when configured.

5. Search and drill-down¶

Goal	Call
Keyword search	`query_engineering_memory(mode=search, query=..., root=<abs>, filters={match_mode: any\\|all})`
Semantic blend	same + `semantic=true` when semantic index is built
One path	`query_engineering_memory(mode=for_path, path=..., root=<abs>)`
Trajectory detail	`query_engineering_memory(mode=trajectory_get, record_id=<trajectory-id>, root=<abs>)`
Trajectory dashboard	`query_engineering_memory(mode=trajectory_dashboard, root=<abs>)`
Playbook	`help(topic=engineering_memory)`

6. Semantic index maintenance¶

When [tool.codeclone.memory.semantic] enabled=true:

manage_engineering_memory(action=rebuild_semantic_index, root=<abs>)

Contract: Semantic search.

7. Trajectory and Experience evidence¶

Scoped get_relevant_memory keeps governed records, trajectory precedents, and advisory Experiences in separate response lanes. Inspect the workflow in Trajectories and Experiences; use promote_experience only when a recurring pattern deserves human review as a draft memory record.