mc-memory-search

# Project Memory Search The auto-loaded `MEMORY.md` is only a curated pointer index — the detail (topic files, the session log, the archive) is NOT in your context. This skill does a ranked search over that full corpus and is the reliable way to recover prior project knowledge mid-task. Relevant memory for your current task is also auto-surfaced in your system context under **"RELEVANT MEMORY"** — check there first. Use this skill when that's absent or insufficient, or when you need to dig into a *different* topic than the task started on. ## How to call it You know your Clayrune project id and port from the SYSTEM block (the same id used by the backlog/process curls). Then: ``` curl -s "http://localhost:5199/api/project/<PROJECT_ID>/memory/search?q=<QUERY>&k=5" ``` - `q` — space-separated keywords (URL-encode spaces as `%20` or `+`). Use the concrete nouns of what you're unsure about: subsystem name, file, feature, decision topic. Terms shorter than 3 chars are ignored. - `k` — max results (default 3; use 5–8 when exploring). Returns a JSON array, ranked best-first: ```json [{"file": "arch_sse_slot_management.md", "score": 7, "snippet": "Chromium caps 6 connections per origin; close SSE on …"}] ``` ## How to use the results 1. The `snippet` is a preview, not the whole story. If a hit looks relevant, **open the referenced file** for full detail: - `<file>.md` → a topic file in the project's memory dir — Read it. - `MEMORY.md#managed` → a session-log e...