memory

Files (1)

• SKILL.md

SKILL.md (first lines)

# memory

Search the memory graph and session memories via `coworker memory query`.

## When to Use

- Searching past session summaries for what was learned or decided
- Finding knowledge cards related to a topic or feature
- Querying the code knowledge graph for relevant files or functions
- Recalling past work before starting a related task

## When NOT to Use

- Extracting new memories from sessions → use /knowledge
- Searching git history → use git log directly
- Searching the web → use web search

## Process

### Step 1: If no query given

Ask the user what they want to search for. Be specific — a few keywords
or a short question works best.

### Step 2: Run the query

```bash
coworker memory query "<user's question>" --mode both --top-k 30
```

**Mode options:**
- `--mode both` (default) — search both the code graph AND session memories
- `--mode graph` — code structure only (functions, files, modules)
- `--mode vector` — session memories only (summaries, knowledge cards)

**Tuning:**
- `--top-k 20` — fewer results, faster
- `--top-k 50` — more results, for thorough searches
- `--min-score 0.5` — stricter relevance filter
- `--budget 4000` — more context per result

### Step 3: Interpret results

The command outputs two tables:
- 🔗 Knowledge Graph — code nodes (functions, files, modules) with
  labels, types, source files, and path weights
- 🧠 Session Memory — memory cards from past sessions with content,
  type, and relevance score

Synthesize what was found and present it clearly to the user.

### stats — Memory graph overview

When the user asks about memory stats or health:

```bash
coworker memory stats
```

Shows: node count by type, edge count by relation, provenance, and edge
health (normal/stale/suppressed).

From cicidi/walter-worker, MIT.

More from cicidi/walter-worker