git:20260729.12a9be5 to git:20260927.31ac526

2 added, 2 removed. Audit B to B.

---

- description: Search the C Brain trunk for notes relevant to a subject, and read the ones that matter. Use when the user asks what they already know about something, wants past decisions on a topic, says "did we solve this before", "what do I have on X", "check my notes", or when a task smells like something already worked out once.

+ description: Search the GreyMatter trunk for notes relevant to a subject, and read the ones that matter. Use when the user asks what they already know about something, wants past decisions on a topic, says "did we solve this before", "what do I have on X", "check my notes", or when a task smells like something already worked out once.

---

# Recall from the trunk

Retrieve what is already known before working something out again.

## How

Run the recall, top 5, and read what comes back:

```bash

brain recall "$ARGUMENTS"

```

If `brain` is not on the PATH, the plugin ships it — call it directly:

`"${CLAUDE_PLUGIN_ROOT}/bin/brain" recall "$ARGUMENTS"`.

Each result gives a score, a note name and a path relative to the trunk

- (`~/.c-brain/trunk`). **Open the ones above the noise floor and actually read

+ (`~/.greymatter/trunk`). **Open the ones above the noise floor and actually read

them** — the ranking is lexical, so it tells you where to look, never what is

true.

## Then

- Say what the notes establish, and cite each one by its path so the user can

check you.

- Say plainly when nothing relevant came back. A confident answer assembled

from three weakly-matching notes is worse than "the trunk has nothing on

this" — it looks like memory and is not.

- Never rewrite a note as a side effect of reading it.

## Scale

Recall holds well to about a thousand notes and degrades past that

(`tests/recall_benchmark.py` publishes the numbers). On a large trunk, prefer

several narrow queries over one broad one.