rh-recalllisted

Recall across every memory store for: **$ARGUMENTS** ``` !`node "$HOME/.claude/scripts/rh-recall.js" "$ARGUMENTS"` ``` ## How to use what comes back The stores answer different questions — read them for what each is good at: | Section | What it tells you | |---|---| | **Concept graph** | Which ideas are linked to this one. Follow `↳ links:` to concepts that would otherwise look far-flung. | | **Project memory** | Standing facts and user rulings for a project. Highest authority — these are deliberate. | | **Shared learnings** | Cross-project techniques and decision rules. | | **Scribe rows + proposals** | Open cleanup/recommendation items. A `↳ proposal:` line is an LLM triage verdict already produced for that row — read it before re-analysing. | | **Past session transcripts** | What was actually said and decided, with session id and project. | | **Oversight logs** | Hook/telemetry events — useful for "when did this start failing". | Then: 1. **Say what you found before acting.** If recall answers the question, cite the file or session id and move on — that is the whole point. Do not silently re-derive it. 2. **Treat memory as dated, not current.** A memory can outlive its evidence, and a fresh session reads memory before it reads the repo. If a hit names a file, flag, or count, re-verify at source before relying on it. 3. **Prefer the source of truth.** The `.md` files are canonical; Postgres is a local index over them and can lag or be absent entirely. 4. *