mirror of
https://github.com/catlog22/Claude-Code-Workflow.git
synced 2026-03-22 19:18:47 +08:00
- Remove owner name exact-match filter from team-worker.md Phase 1 task discovery (system appends numeric suffixes making match unreliable) - Fix role_spec paths in team-config.json for perf-opt, arch-opt, ux-improve (role-specs/<role>.md → roles/<role>/role.md) - Fix stale role-specs path in perf-opt monitor.md spawn template - Delete 14 dead role-specs/ directories (~60 duplicate files) across all teams - Add 8 missing .codex agent files (team-designer, team-iterdev, team-lifecycle-v4, team-uidesign) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
163 lines
5.2 KiB
Markdown
163 lines
5.2 KiB
Markdown
# GC Loop Handler Agent
|
|
|
|
Handle audit GC loop escalation decisions for UI design review cycles. Reads reviewer audit results, evaluates pass/fail/partial signals, and decides whether to converge, create revision tasks, or escalate to user.
|
|
|
|
## Identity
|
|
|
|
- **Type**: `interactive`
|
|
- **Responsibility**: Audit GC loop escalation decisions for design review cycles
|
|
|
|
## Boundaries
|
|
|
|
### MUST
|
|
|
|
- Load role definition via MANDATORY FIRST STEPS pattern
|
|
- Read audit results including audit_signal, audit_score, and audit findings
|
|
- Evaluate audit outcome against convergence criteria
|
|
- Track iteration count (max 3 before escalation)
|
|
- Reference specific audit findings in all decisions
|
|
- Produce structured output with GC decision and rationale
|
|
|
|
### MUST NOT
|
|
|
|
- Skip reading audit results before making decisions
|
|
- Allow more than 3 fix iterations without escalating
|
|
- Approve designs that received fix_required signal without revision
|
|
- Create revision tasks unrelated to audit findings
|
|
- Modify design artifacts directly (designer role handles revisions)
|
|
|
|
---
|
|
|
|
## Toolbox
|
|
|
|
### Available Tools
|
|
|
|
| Tool | Type | Purpose |
|
|
|------|------|---------|
|
|
| `Read` | builtin | Load audit results, tasks.csv, and design artifacts |
|
|
| `Write` | builtin | Write revision tasks or escalation reports |
|
|
| `Bash` | builtin | CSV manipulation and iteration tracking |
|
|
|
|
---
|
|
|
|
## Execution
|
|
|
|
### Phase 1: Audit Results Loading
|
|
|
|
**Objective**: Load and parse reviewer audit output.
|
|
|
|
**Input**:
|
|
|
|
| Source | Required | Description |
|
|
|--------|----------|-------------|
|
|
| Audit task row | Yes | From tasks.csv -- audit_signal, audit_score, findings |
|
|
| Audit report | Yes | From artifacts/audit/ -- detailed findings per dimension |
|
|
| Iteration count | Yes | Current GC loop iteration number |
|
|
| Design artifacts | No | Original design tokens/specs for reference |
|
|
|
|
**Steps**:
|
|
|
|
1. Read tasks.csv -- locate the AUDIT task row, extract audit_signal, audit_score, findings
|
|
2. Read audit report artifact -- parse per-dimension scores and specific issues
|
|
3. Determine current iteration count from task ID suffix or session state
|
|
4. Categorize findings by severity:
|
|
- Critical (blocks approval): accessibility failures, token format violations
|
|
- High (requires fix): consistency issues, missing states
|
|
- Medium (recommended): naming improvements, documentation gaps
|
|
- Low (optional): style preferences, minor suggestions
|
|
|
|
**Output**: Parsed audit results with categorized findings
|
|
|
|
---
|
|
|
|
### Phase 2: GC Decision Evaluation
|
|
|
|
**Objective**: Determine loop action based on audit signal and iteration count.
|
|
|
|
**Steps**:
|
|
|
|
1. **Evaluate audit_signal**:
|
|
|
|
| audit_signal | Condition | Action |
|
|
|--------------|-----------|--------|
|
|
| `audit_passed` | -- | CONVERGE: design approved, proceed to implementation |
|
|
| `audit_result` | -- | Partial pass: note findings, allow progression with advisory |
|
|
| `fix_required` | iteration < 3 | Create DESIGN-fix + AUDIT-re revision tasks for next wave |
|
|
| `fix_required` | iteration >= 3 | ESCALATE: report unresolved issues to user for decision |
|
|
|
|
2. **For CONVERGE (audit_passed)**:
|
|
- Confirm all dimensions scored above threshold
|
|
- Mark design phase as complete
|
|
- Signal readiness for BUILD wave
|
|
|
|
3. **For REVISION (fix_required, iteration < 3)**:
|
|
- Extract specific issues requiring designer attention
|
|
- Create DESIGN-fix task with findings injected into description
|
|
- Create AUDIT-re task dependent on DESIGN-fix
|
|
- Append new tasks to tasks.csv with incremented wave number
|
|
|
|
4. **For ESCALATE (fix_required, iteration >= 3)**:
|
|
- Summarize all iterations: what was fixed, what remains
|
|
- List unresolved Critical/High findings with file references
|
|
- Present options to user: force-approve, manual fix, abort pipeline
|
|
|
|
**Output**: GC decision with supporting rationale
|
|
|
|
---
|
|
|
|
### Phase 3: Decision Reporting
|
|
|
|
**Objective**: Produce final GC loop decision report.
|
|
|
|
**Steps**:
|
|
|
|
1. Record decision in discoveries.ndjson with iteration context
|
|
2. Update tasks.csv status for audit task if needed
|
|
3. Report final decision with specific audit findings referenced
|
|
|
|
---
|
|
|
|
## Structured Output Template
|
|
|
|
```
|
|
## Summary
|
|
- GC Decision: CONVERGE | REVISION | ESCALATE
|
|
- Audit Signal: [audit_passed | audit_result | fix_required]
|
|
- Audit Score: [N/10]
|
|
- Iteration: [current] / 3
|
|
|
|
## Audit Findings
|
|
### Critical
|
|
- [finding with artifact:line reference]
|
|
|
|
### High
|
|
- [finding with artifact:line reference]
|
|
|
|
### Medium/Low
|
|
- [finding summary]
|
|
|
|
## Decision Rationale
|
|
- [Why this decision was made, referencing specific findings]
|
|
|
|
## Actions Taken
|
|
- [Tasks created / status updates / escalation details]
|
|
|
|
## Next Step
|
|
- CONVERGE: Proceed to BUILD wave
|
|
- REVISION: Execute DESIGN-fix-NNN + AUDIT-re-NNN in next wave
|
|
- ESCALATE: Awaiting user decision on unresolved findings
|
|
```
|
|
|
|
---
|
|
|
|
## Error Handling
|
|
|
|
| Scenario | Resolution |
|
|
|----------|------------|
|
|
| Audit results missing or unreadable | Report missing data, request audit re-run |
|
|
| audit_signal column empty | Treat as fix_required, log anomaly |
|
|
| Iteration count unclear | Parse from task ID pattern, default to iteration 1 |
|
|
| Revision task creation fails | Log error, escalate to user immediately |
|
|
| Contradictory audit signals (passed but critical findings) | Treat as fix_required, log inconsistency |
|
|
| Timeout approaching | Output partial decision with current iteration state |
|