bun -e "
import { loadDevBuddyConfig, getProviderType } from '${CLAUDE_PLUGIN_ROOT}/scripts/pipeline-config.ts';
const config = loadDevBuddyConfig();
const stageType = '{plan-review or code-review}';
const stage = config.stages[stageType];
const executors = stage.executors.map(exec => ({
  ...exec,
  providerType: getProviderType(exec.preset)
}));
console.log(JSON.stringify({ executors, max_iterations: config.max_iterations }));
"

You are executing the PLAN REVIEW stage.
Your system prompt name is: {executor.system_prompt}
Your model is: {executor.model}
Set revision_number to: {revision_number}

PESSIMISTIC-FIRST: Assume NOTHING in this plan will work as described.
For each step:
1. What specific test would fail if this step has a bug? Reference test IDs from the TDD Test Plan.
2. Is this step truly one architectural unit? If it touches multiple modules, flag it.
3. Does it reuse existing code? If it creates new code where existing works, flag it.
4. Is the rollback specific? "Revert changes" is not acceptable.

REQUIREMENTS:
{extracted ACs}

TDD TEST PLAN:
{extracted test IDs with AC mappings}

RISK REGISTRY:
{extracted risks — flag unacknowledged}

IMPLEMENTATION STEPS:
{extracted steps with AC/test mappings}

Review and write output to {TMPDIR}/.vcp/oneshot/review-{RAND}-{i}.json

You are executing the CODE REVIEW stage.
Your system prompt name is: {executor.system_prompt}
Your model is: {executor.model}
Set revision_number to: {revision_number}

PESSIMISTIC-FIRST: Assume every line of changed code has a bug. Find them.
For each AC:
1. Find the specific code path that implements it — cite file:line
2. Trace input → processing → output through that path
3. Identify what would break if any step fails

REQUIREMENTS:
{extracted ACs}

TDD TEST PLAN:
{extracted tests}

GIT DIFF:
{actual code changes}

Review and write output to {TMPDIR}/.vcp/oneshot/review-{RAND}-{i}.json

## Plan Review Record

```json
{
  "id": "review-YYYYMMDD-HHMMSS",
  "reviewer": "synthesizer system_prompt name (or first reviewer if single)",
  "model": "synthesizer model (or first reviewer model if single)",
  "revision_number": 1,
  "reviewers": [{"name": "{system_prompt}", "model": "{model}"}],
  "status": "approved|needs_changes|needs_clarification|rejected",
  "summary": "2-3 sentence consolidated summary",
  "needs_clarification": false,
  "clarification_questions": [],
  "findings": [{all must_fix and advisory findings from all reviewers, deduplicated}],
  "requirements_coverage": {
    "mapping": [{"ac_id": "AC-1", "steps": ["step 1"], "tests": ["UT-1"]}],
    "acs_without_steps": [],
    "acs_without_tests": [],
    "steps_without_ac": [],
    "steps_without_tests": [],
    "risks_unacknowledged": []
  },
  "reviewed_at": "{ISO8601}"
}

**For code review, append `## Code Review Record` and update `## Sign-off`:**
```markdown
## Code Review Record

```json
{
  "id": "code-review-YYYYMMDD-HHMMSS",
  "reviewer": "synthesizer system_prompt name",
  "model": "synthesizer model",
  "revision_number": 1,
  "reviewers": [{"name": "{system_prompt}", "model": "{model}"}],
  "status": "approved|needs_changes|needs_clarification|rejected",
  "summary": "2-3 sentence consolidated summary",
  "needs_clarification": false,
  "clarification_questions": [],
  "acceptance_criteria_verification": {
    "total": 6, "verified": 5, "missing": ["AC-3"],
    "details": [{"ac_id": "AC-1", "status": "IMPLEMENTED", "evidence": "src/auth.ts:42", "notes": ""}]
  },
  "findings": [{all must_fix and advisory findings, deduplicated}],
  "checklist": {
    "security_owasp": "PASS|WARN|FAIL",
    "error_handling": "PASS|WARN|FAIL",
    "resource_management": "PASS|WARN|FAIL",
    "configuration": "PASS|WARN|FAIL",
    "code_quality": "PASS|WARN|FAIL",
    "concurrency": "PASS|WARN|FAIL|N/A",
    "logging": "PASS|WARN|FAIL",
    "dependencies": "PASS|WARN|FAIL",
    "api_design": "PASS|WARN|FAIL|N/A",
    "backward_compatibility": "PASS|WARN|FAIL|N/A",
    "testing": "PASS|WARN|FAIL",
    "over_engineering": "PASS|WARN|FAIL"
  },
  "reviewed_at": "{ISO8601}"
}

{
  "requirements_approved": "{date}",
  "plan_approved": "{date}",
  "implementation_complete": "{date}",
  "code_review_passed": "{date or null}"
}

**If this is a re-review (revision > 1):** Replace the existing review record section instead of appending a second one.

---

## Step 11: Review → Repair → Re-Review Loop

If aggregate status is `needs_changes` and iterations < max_iterations:

1. Extract `must_fix` findings from the review record
2. **For plan review:** Dispatch `/dev-buddy-plan` via Skill tool (it will read the review findings from the plan file and re-plan)
3. **For code review:** Dispatch `/dev-buddy-implement` via Skill tool (it will read the review findings and fix)
4. After repair completes, increment revision_number
5. Delete the current review temp files
6. Re-dispatch ALL reviewers with new revision_number
7. Return to Step 8

If iterations exhausted → update status to `rejected` in the review record.

---

## Step 12: Cleanup and Report

1. Remove temp files: `rm -f "{TMPDIR}/.vcp/oneshot/review-{RAND}-"*`
2. Present to user:
   - Overall status
   - Number of findings (must_fix vs advisory)
   - AC coverage / verification status
3. If approved, suggest next step:
   - Plan review approved → `/dev-buddy-implement`
   - Code review approved → done

---

## Error Handling

| Scenario | Action |
|----------|--------|
| Required plan file sections missing | Tell user which stage to run first |
| All reviewers fail | Report error to user |
| Single reviewer fails | Continue with remaining |
| Max iterations exceeded | Set status to rejected, report to user |
| Clarification answer in plan file | Resolve without asking user |