claude-hangar /pattern-extractor — Learn from History
Analyzes the command patterns captured by the continuous-learning hook and extracts actionable insights.
What It Does
- Reads pattern logs from
~/.claude/.patterns/ - Identifies recurring success/failure patterns
- Extracts reusable workflows (commands that consistently succeed)
- Flags anti-patterns (commands that consistently fail)
- Generates a summary with recommendations
Usage
/pattern-extractor # Analyze all patterns
/pattern-extractor last-week # Analyze last 7 days
/pattern-extractor project # Analyze current project only
Analysis Process
Step 1: Load Pattern Data
Read all .jsonl files from ~/.claude/.patterns/:
cat ~/.claude/.patterns/session-*.jsonl
Step 2: Categorize Patterns
Group commands by:
- Recovery patterns: Command that succeeded after a similar command failed
- Workflow patterns: Sequences of commands that appear together
- Failure patterns: Commands that consistently fail in certain contexts
- Tool preferences: Which tools are used most for which tasks
Step 3: Generate Insights
For each pattern category, produce:
## Recovery Patterns
| Failed Command | Successful Recovery | Frequency |
|----------------|---------------------|-----------|
| npm run build | npm ci && npm run build | 5 times |
## Workflow Patterns
| Workflow | Steps | Frequency |
|----------|-------|-----------|
| TDD cycle | test → edit → test | 23 times |
## Anti-Patterns (Avoid)
| Command | Failure Rate | Recommendation |
|---------|-------------|----------------|
| git push -f | 80% blocked | Use regular push |
Step 4: Save Insights
Write analysis to ~/.claude/.patterns/insights-YYYY-MM-DD.md.
If insights reveal a reusable pattern, suggest creating a skill for it.
Output
Summary with:
- Top 5 recovery patterns
- Top 5 workflow patterns
- Top 5 anti-patterns
- Recommendation: patterns worth formalizing as skills