| name | review-agent-work |
| description | Reviews and validates work completed by another agent against the original plan and Kubernaut project guidelines. Checks TDD compliance, business requirement mapping, integration patterns, and code quality. Use when verifying agent work, checking plan adherence, or validating implementation completeness. |
Agent Work Review
Validates completed work against the execution plan and Kubernaut core development standards to ensure alignment and quality.
When to Use
- After another agent completes implementation work
- Before marking tasks as complete
- When verifying plan adherence
- Before committing or submitting changes
Review Process
Step 1: Understand the Context
MANDATORY: Achieve 100% confidence in understanding what needs to be reviewed.
- Read the plan/task description: Understand the original goals, acceptance criteria, and approach
- Identify changed files: Review what was actually modified
- Check for TODO items: Look for any task tracking that indicates scope
CRITICAL: Ask for user input when ANYTHING is unknown, unclear, or uncertain:
- The plan is unclear or missing key details
- Changed files don't match expected scope
- There are conflicting requirements
- Business requirements are ambiguous
- Technical approach is questionable
- Expected behavior is not documented
- You're unsure about ANY aspect of what to validate
DO NOT proceed to Step 2 until you have 100% confidence in what needs to be checked.
DO NOT make assumptions. When in doubt, ASK.
Step 2: Execute Review Checklist
IMPORTANT: Stop and ask for input if you encounter ANY uncertainty during the review:
- You cannot determine if a requirement is met
- Expected behavior is unclear
- Test coverage adequacy is ambiguous
- Integration points are uncertain
- You're making assumptions to complete the checklist
When uncertain: PAUSE the review and ask the user for clarification.
Copy this checklist and verify each item:
## Review Checklist
### Plan Alignment
- [ ] All planned tasks/features implemented
- [ ] No unplanned changes or scope creep
- [ ] Original acceptance criteria met
- [ ] Edge cases from plan addressed
### Business Requirements (Kubernaut Core)
- [ ] Business requirement(s) identified (BR-[CATEGORY]-[NUMBER])
- [ ] Code changes map to documented business needs
- [ ] No speculative or "nice to have" code without BR backing
### TDD Compliance
- [ ] Tests written first (RED phase evidence)
- [ ] Tests validate **business outcomes/behavior**, NOT implementation details
- [ ] Minimal implementation passes tests (GREEN phase)
- [ ] Code enhanced/refactored (REFACTOR phase)
- [ ] Ginkgo/Gomega BDD framework used (not standard Go testing)
- [ ] Test descriptions include BR or test scenario IDs
- [ ] NO Skip() or pending tests present
### Code Quality
- [ ] All errors handled and logged
- [ ] No new lint or compilation errors
- [ ] Structured types used (avoid `any`/`interface{}`)
- [ ] Error wrapping includes context
- [ ] No unused/dead code introduced
### Integration (Kubernaut Critical)
- [ ] Business code integrated with main applications (cmd/)
- [ ] Type definitions validated before field access
- [ ] No orphaned components or unused code
- [ ] Existing patterns enhanced (not reinvented)
### Build & Test Validation
- [ ] Code builds without errors (`go build ./...`)
- [ ] Tests pass (`go test ./...`)
- [ ] No refactoring artifacts left behind (old field names, types)
### Confidence Assessment Provided
- [ ] Confidence percentage provided (60-100%)
- [ ] Justification includes risks and assumptions
- [ ] Validation approach documented
Step 3: Generate Review Report
Provide output in this format:
# Agent Work Review Results
## Context
- **Task/Plan**: [Brief description]
- **Files Changed**: [Count and key files]
- **Reviewer Confidence**: [Percentage]%
## Checklist Results
[Pass/Fail status for each section from checklist above]
### Plan Alignment: ✅ PASS / ❌ FAIL
- [Specific findings]
### Business Requirements: ✅ PASS / ❌ FAIL
- [Specific findings]
### TDD Compliance: ✅ PASS / ❌ FAIL
- [Specific findings]
### Code Quality: ✅ PASS / ⚠️ WARNINGS / ❌ FAIL
- [Specific findings]
### Integration: ✅ PASS / ❌ FAIL
- [Specific findings]
### Build & Test Validation: ✅ PASS / ❌ FAIL
- [Specific findings]
## Overall Assessment
**Status**: ✅ APPROVED / ⚠️ APPROVED WITH WARNINGS / ❌ REQUIRES CHANGES
**Blocking Issues** (must fix):
- [List any critical issues]
**Warnings** (should address):
- [List any non-critical concerns]
**Confidence Score**: [60-100]%
**Rationale**: [Explain confidence level and any uncertainties]
## Next Steps
- [What needs to happen next - fix issues, proceed to commit, etc.]
Quick Reference: Kubernaut Core Rules
TDD RED-GREEN-REFACTOR
- RED: Write failing tests defining business contract
- GREEN: Minimal implementation + mandatory main app integration
- REFACTOR: Enhance existing code with sophisticated logic
Business Integration
Business code MUST be integrated in main applications (cmd/):
grep -r "[NewComponent]" cmd/ --include="*.go"
Common Violations to Check
- ❌ Tests using Skip() to avoid failures
- ❌ Implementation without tests (RED phase skipped)
- ❌ Tests validating implementation logic instead of business behavior/outcomes
- ❌ Sophisticated logic in GREEN phase (should be in REFACTOR)
- ❌ New types not integrated with main apps
- ❌ Struct field references without type validation
- ❌ Refactoring without build validation
Tool Usage Examples
Check for business requirement mapping
rg "BR-[A-Z]+-[0-9]+" [test_file] [implementation_file]
Verify main application integration
grep -r "ComponentName\|NewType" cmd/ --include="*.go"
Validate build after refactoring
go build ./...
go test ./... -run=^$ -timeout=30s
Check for Skip() usage
rg "Skip\(\)" test/ --include="*_test.go"
Escalation Criteria
Stop the review and escalate to the user if:
- Plan is unclear or incomplete - Cannot determine acceptance criteria
- Major architectural changes - Scope significantly exceeds plan
- Critical safety issues - Security vulnerabilities, data loss risks
- Complete TDD violation - No tests written, all tests skipped
- Orphaned business code - New components not integrated anywhere
- Confidence below 80% - Uncertainty about correctness or completeness
- ANY uncertainty or ambiguity - When you need input to complete the review
When to Ask for User Input
MANDATORY: Ask for user input when:
During Initial Assessment (Step 1)
- Plan/task description is vague or incomplete
- Acceptance criteria are not clearly defined
- Business requirements are ambiguous or missing
- Scope of work is unclear
During Review Execution (Step 2)
- You cannot verify a checklist item without making assumptions
- Expected behavior is not documented
- Test coverage adequacy is subjective/unclear
- Integration patterns are uncertain
- You're unsure if something meets the standard
During Report Generation (Step 3)
- Confidence is below 80% and you need clarification
- You're uncertain about severity (blocking vs. warning)
- Recommendations depend on project context you don't have
General Rule
If you find yourself thinking "I assume..." or "Probably..." → STOP and ASK
Examples of when to ask:
- "I assume this component will be used in X service, but I don't see the integration. Should I flag this?"
- "The test coverage is 65%. Is this acceptable given the component type?"
- "This pattern differs from existing code. Is this intentional or should I flag it?"
- "I'm not sure if this qualifies as a critical security issue. Can you confirm?"
DO NOT:
- Make assumptions and proceed
- Guess at user intent
- Apply arbitrary thresholds without context
- Complete review with significant uncertainties
ALWAYS prefer asking over guessing.
Anti-Patterns in Reviews
❌ Rubber-Stamping
Don't just check boxes without validating. Actually verify each item.
❌ Assuming Without Checking
Don't assume main app integration exists. Use grep to verify.
❌ Ignoring Failed Checks
Don't mark APPROVED if critical items fail. Be honest about blocking issues.
❌ Vague Feedback
Don't say "looks good" or "needs work". Provide specific, actionable findings.
❌ Making Assumptions
Don't proceed with assumptions when uncertain. Ask for clarification instead.
Success Criteria
A successful review:
- ✅ Provides clear pass/fail status for each area
- ✅ Identifies specific blocking issues vs. warnings
- ✅ Includes confidence assessment with rationale
- ✅ Gives actionable next steps
- ✅ Catches critical Kubernaut guideline violations
- ✅ Verifies plan adherence completely