# Pattern Mining Workflow
Automatically extract and promote frequently-used patterns from your sessions.
## Scenario
You repeatedly use certain imports, commands, or code patterns. The mining pipeline detects these and suggests promoting them to hot cache.
## How Mining Works
```
1. Claude's responses are logged (via Stop hook)
2. run_mining() extracts patterns:
- Import statements
- Project facts ("This project uses X")
- Code patterns (function signatures)
- Shell commands
3. Patterns seen 3+ times become promotion candidates
4. You approve candidates to promote to hot cache
```
## Step 1: Enable Auto-Logging
Install the hook (see README):
```json
// ~/.claude/settings.json
{
"hooks": {
"Stop": [{
"hooks": [{
"type": "command",
"command": "/path/to/memory-mcp/hooks/memory-log-response.sh"
}]
}]
}
}
```
## Step 2: Work Normally
Over several sessions, Claude generates responses containing patterns:
```
Claude: "Here's how to run the tests:
```bash
uv run pytest tests/ -v
```"
Claude: "Let me add the import:
```python
from fastapi import FastAPI, HTTPException, Depends
```"
Claude: "This project uses SQLAlchemy 2.0 with async support..."
```
These get logged automatically.
## Step 3: Run Mining
Periodically run mining to extract patterns:
```
You: "Run pattern mining"
Claude: [calls run_mining(hours=24)]
→ Processed 47 outputs, found 12 patterns
You: "What patterns are ready for promotion?"
Claude: [calls review_candidates()]
→ [
{id: 1, pattern: "from fastapi import FastAPI, HTTPException, Depends",
type: "import", occurrences: 5},
{id: 2, pattern: "uv run pytest tests/ -v", type: "command", occurrences: 4},
{id: 3, pattern: "This project uses SQLAlchemy 2.0", type: "fact", occurrences: 3}
]
```
## Step 4: Approve or Reject
```
You: "Approve the FastAPI import pattern"
Claude: [calls approve_candidate(1)]
→ Pattern promoted to hot cache as memory #15
You: "Reject the pytest command - I use different flags now"
Claude: [calls reject_candidate(2)]
→ Pattern rejected and removed from candidates
```
## Step 5: Check Mining Status
```
You: "Show mining status"
Claude: [calls mining_status()]
→ {
outputs_logged: 234,
patterns_extracted: 45,
candidates_pending: 8,
promoted_this_week: 3,
log_retention_days: 7
}
```
## Pattern Types
| Type | What's Extracted | Example |
|------|------------------|---------|
| `import` | Python imports | `from pydantic import BaseModel` |
| `command` | Shell commands | `uv run pytest -v` |
| `fact` | Project facts | "This project uses PostgreSQL" |
| `code` | Function signatures | `def process_data(df: pd.DataFrame)` |
## CLI Mining
Run mining from command line:
```bash
# Process last 24 hours
uv run memory-mcp-cli run-mining --hours 24
# Process last week
uv run memory-mcp-cli run-mining --hours 168
# Get JSON output
uv run memory-mcp-cli --json run-mining
```
## Tips
- **Run mining daily**: `run_mining(hours=24)` catches new patterns
- **Review candidates weekly**: Don't let the queue grow too large
- **Reject outdated patterns**: If your workflow changed, reject old patterns
- **Check occurrences**: Higher count = more likely to be useful
