Scrutiny Validator

Run code scrutiny validation during mission checkpoints. Spawns review subagents, synthesizes results, and produces validation reports.

Arguments

$ARGUMENTS

• <milestone>: Validate a specific milestone (e.g., bootstrap, core)
• --all: Validate all milestones
• --strict: Fail on warnings as well as errors

When to Use This Skill

Use during Mission Control validation checkpoints:

• After completing a milestone
• Before sealing a milestone
• During maestro validation show review
• When maestro:agent-base instructs final validation

Required Skills

• maestro:agent-base - For startup/cleanup procedures

Work Procedure

Step 1: Load Validation State

Check current validation status:

maestro validation show

This displays:

• Active milestone
• Completed validations
• Pending validations
• Blockers

Step 2: Determine Validation Scope

If milestone specified:

• Validate only features in that milestone

If `--all` specified:

• Validate all completed features across all milestones

Default (no args):

• Validate current active milestone

Step 3: Identify Features to Validate

Query features needing validation:

maestro feature list --mission <id> --milestone <name>

Filter for features with status implemented or pending_validation.

Step 4: Spawn Review Subagents

For each feature requiring validation:

1. Prepare review context:

• Feature specification
• Implementation changes (git diff)
• Test results
• Verification steps from feature definition

1. Spawn scrutiny-feature-reviewer:

• Pass feature context
• Pass validation contract assertions
• Request structured review output

Step 5: Synthesize Review Results

Collect and synthesize all subagent reports:

interface ScrutinySynthesis {
  milestone: string;
  featuresReviewed: number;
  featuresPassed: number;
  featuresFailed: number;
  blockers: Array<{
    featureId: string;
    severity: "blocking" | "warning";
    issue: string;
  }>;
  recommendations: string[];
}

Step 6: Update Validation State

Record results in validation state:

maestro validation update --milestone <name> --status <passed|failed>

Next step: return results to maestro:conduct. If status is passed, conduct proceeds to maestro:user-testing-validator (if applicable) or directly to maestro milestone seal. If status is failed, conduct re-opens the affected features for agent re-dispatch. Do not seal the milestone yourself -- that is the conductor's gate.

Step 7: Produce Validation Report

Generate human-readable and JSON reports:

Human format:

=== Scrutiny Validation: <milestone> ===

Features Reviewed: 5
Passed: 4
Failed: 1

Blockers:
- feat-003: Missing error handling (blocking)

Recommendations:
1. Add input validation for edge cases
2. Increase test coverage for error paths

Status: FAILED (1 blocking issue)

JSON format (with --json):

{
  "milestone": "core",
  "summary": {
    "featuresReviewed": 5,
    "featuresPassed": 4,
    "featuresFailed": 1
  },
  "blockers": [...],
  "recommendations": [...],
  "status": "failed"
}

Review Dimensions

Each feature is scrutinized on:

Validation States

proposed → approved → in_progress → implemented → validated
                                              ↓
                                        validation_failed → in_progress

States map to CLI commands:

• maestro feature approve → approved
• Agent completes → implemented
• maestro:scrutiny-validator passes → validated
• maestro:scrutiny-validator fails → validation_failed

Related Commands

Integration with Milestones

Scrutiny validation runs automatically when:

1. All features in a milestone are implemented
2. Before maestro milestone seal can complete
3. During maestro validation show if pending

Best Practices

1. Validate incrementally - Don't wait until the end
2. Fix blockers immediately - Don't seal with blockers
3. Document warnings - Even non-blocking issues should be tracked
4. Use strict mode for critical missions - --strict treats warnings as failures
5. Save checkpoint after validation - maestro checkpoint save