aiwg
Version:
Cognitive architecture for AI-augmented software development with structured memory, ensemble validation, and closed-loop correction. FAIR-aligned artifacts, 84% cost reduction via human-in-the-loop, standards adopted by 100+ organizations.
143 lines (102 loc) • 3.58 kB
Markdown
# AIWG Hooks
Claude Code hook templates for workflow tracing, permissions, and session management.
## Research Foundation
- **REF-001**: BP-6 - Production workflows require trace capture for observability
- **REF-002**: Archetype 4 - Recovery requires execution history
## Installation
```bash
aiwg install aiwg-hooks
```
Or manual:
```bash
cp -r aiwg-hooks/hooks/* ~/.config/claude/hooks/
```
## Hooks
### aiwg-trace.js
Captures multi-agent workflow execution traces.
**Events Captured:**
- `SubagentStart` - Agent spawn with metadata (type, model, tools)
- `SubagentStop` - Completion with outcome, duration, transcript path
**Configuration:**
Add to `.claude/settings.local.json`:
```json
{
"hooks": {
"SubagentStart": ["node", "~/.config/claude/hooks/aiwg-trace.js", "start"],
"SubagentStop": ["node", "~/.config/claude/hooks/aiwg-trace.js", "stop"]
}
}
```
**Environment Variables:**
| Variable | Default | Description |
|----------|---------|-------------|
| `AIWG_TRACE_DIR` | `.aiwg/traces` | Trace output directory |
| `AIWG_TRACE_FILE` | `current-trace.jsonl` | Trace filename |
| `AIWG_TRACE_VERBOSE` | `false` | Log to stderr |
| `AIWG_WORKFLOW_ID` | (auto) | Workflow identifier |
**Trace Format (JSONL):**
```json
{"timestamp":"2025-01-15T10:30:00.000Z","type":"agent_start","agent_id":"abc123","subagent_type":"security-architect","model":"sonnet"}
{"timestamp":"2025-01-15T10:30:45.000Z","type":"agent_stop","agent_id":"abc123","outcome":"success","duration_ms":45000}
```
## Scripts
### trace-viewer.mjs
View and analyze workflow traces.
```bash
# View current trace as tree
aiwg trace-view
# View as timeline
aiwg trace-view --format timeline
# Filter by agent type
aiwg trace-view --filter security
# View specific trace file
aiwg trace-view --trace .aiwg/traces/2025-01-15.jsonl
# Export as JSON
aiwg trace-view --format json > trace.json
```
**Output Formats:**
**Tree** (default):
```
✓ architecture-designer [opus] (45000ms)
✓ security-architect [sonnet] (12000ms)
✓ test-architect [sonnet] (8000ms)
✓ technical-writer [haiku] (5000ms)
✓ documentation-synthesizer [sonnet] (15000ms)
```
**Timeline:**
```
| Time | Event | Agent | Details |
|----------|-------|--------------------|--------------------|
| 10:30:00 | START | architecture-designer | model=opus |
| 10:30:05 | START | security-architect | model=sonnet |
| 10:30:17 | STOP | security-architect | success (12000ms) |
```
## Trace Directory Structure
```
.aiwg/traces/
├── current-trace.jsonl # Active workflow trace
├── 2025-01-15-workflow-abc.jsonl # Archived traces
└── transcripts/ # Captured agent transcripts
├── agent-abc123.json
└── agent-def456.json
```
## Integration with Workflows
Set workflow ID for correlated traces:
```bash
export AIWG_WORKFLOW_ID="flow-security-$(date +%Y%m%d-%H%M%S)"
/flow-security-review-cycle
```
Archive trace after workflow:
```bash
mv .aiwg/traces/current-trace.jsonl .aiwg/traces/$AIWG_WORKFLOW_ID.jsonl
```
## Observability Benefits
1. **Debugging**: Trace agent interactions to identify failures
2. **Performance**: Identify slow agents or bottlenecks
3. **Recovery**: Use traces to replay failed workflows
4. **Audit**: Compliance trail of agent decisions
5. **Optimization**: Analyze parallel execution patterns
## Related
- `prompts/reliability/resilience.md` - Recovery protocol using traces
- `flow-*.md` - Workflow commands that generate traces
- Claude Code hooks documentation