Quality Gates - Approval Criteria Standardization

Purpose

Standardize approval/rejection decisions across all reviewer agents (plan-reviewer, code-reviewer, documentation-reviewer, test-reviewer) using consistent criteria.

<!-- CUSTOMIZE: Adjust language based on project conventions -->

4-Tier Approval System

1. ✅ APPROVED (Proceed Immediately)

Definition: Meets all required criteria, ready to proceed to next step immediately

Criteria:

• All required requirements met
• No architecture/security/performance issues
• Code quality standards met
• Only optional improvements exist (non-blocking)

Output Format:

## Overall Assessment
✅ **APPROVED** - Ready for next step

## Critical Issues
None

## Important Improvements
None

## Suggestions
- [Optional improvement 1]
- [Optional improvement 2]

Conductor Action: Immediately proceed to next step

---

2. ⚠️ APPROVED_WITH_CONDITIONS (Conditional Approval)

Definition: Mostly passed but requires specific conditions to be met before proceeding

Criteria:

• Core functionality complete
• Important but not immediately blocking issues exist
• Conditions can be quickly verified by conductor

Condition Examples:

• "Proceed after confirming security issue #42 resolved"
• "Proceed after performance test results show < {{X}}ms"
• "Proceed after user approval"

Output Format:

## Overall Assessment
⚠️ **APPROVED_WITH_CONDITIONS**

## Conditions for Approval
1. [Condition 1]: [Verification method]
2. [Condition 2]: [Verification method]

## Critical Issues
None

## Important Improvements
- [Issue requiring condition]

Conductor Action:

1. Verify conditions
2. If conditions met → Proceed to next step
3. If conditions not met → Treat as NEEDS_REVISION

---

3. 🔄 NEEDS_REVISION (Return to Previous Step)

Definition: Required criteria not met, rework needed

Criteria:

• Required requirements not met
• Architecture issues requiring redesign
• Security vulnerabilities found
• Performance criteria not met
• Plan and implementation mismatch

Output Format:

## Overall Assessment
🔄 **NEEDS_REVISION**

## Critical Issues (Must Fix)
1. [Issue 1]: [Why critical] → [Required action]
2. [Issue 2]: [Why critical] → [Required action]

## Important Improvements (Should Fix)
- [Improvement 1]

## Recommended Actions
1. [Step to take 1]
2. [Step to take 2]

Conductor Action:

1. Request rework from previous step agent
2. Re-review after critical issues resolved
3. Prevent infinite loop: Escalate to user after 3 iterations

---

4. 🚫 REJECTED (Stop Workflow)

Definition: Fundamental problems requiring workflow stop and user decision

Criteria:

• Fundamental architecture flaws (complete redesign needed)
• Plan itself has problems (requirements redefinition needed)
• Security risk too high (alternative approach needed)
• Technically infeasible

Output Format:

## Overall Assessment
🚫 **REJECTED** - Workflow requires user decision

## Blocking Issues
1. [Fundamental problem 1]: [Why blocking] → [Alternative approach needed]
2. [Fundamental problem 2]: [Why blocking] → [Alternative approach needed]

## Recommended Next Steps
1. [User decision needed on X]
2. [Consider alternative approach Y]
3. [Escalate to senior architect]

Conductor Action:

1. Stop workflow immediately
2. Report situation to user
3. Wait for user decision:
   • Complete plan revision
   • Select alternative approach
   • Reduce feature scope
   • Stop workflow

---

Reviewer-Specific Application Guide

plan-reviewer (Step 2)

APPROVED:

• Phase breakdown clear
• Dependency analysis complete
• Risks identified
• Code minimization verified

APPROVED_WITH_CONDITIONS:

• "Proceed after user selects architecture option A/B"
• "Proceed after performance test strategy confirmed"

NEEDS_REVISION:

• Insufficient phase breakdown
• Code bloat concern (provide alternatives)
• Security risks not assessed

REJECTED:

• Fundamental architecture flaws
• Requirements ambiguous (redefinition needed)

---

code-reviewer (Step 4)

APPROVED:

• Plan compliance
• Architecture principles met
• No security issues
• Excellent code quality

APPROVED_WITH_CONDITIONS:

• "Proceed after confirming TODO comments removed"
• "Proceed after confirming type hints added"

NEEDS_REVISION:

• Security vulnerabilities (RCE, SQL Injection)
• Architecture violations (SSOT, DRY)
• Performance issues (N+1 query)
• Plan non-compliance

REJECTED:

• Implementation completely different from plan
• Unfixable security design flaws

---

documentation-reviewer (Step 6)

<!-- CUSTOMIZE: Set MAX_DOC_LINES based on project standards -->

APPROVED:

• Under {{MAX_DOC_LINES}} lines
• Clear and concise
• Matches code
• Tag consistency

APPROVED_WITH_CONDITIONS:

• "Proceed after reducing Known Issues to < 5 lines"
• "Proceed after updating example code"

NEEDS_REVISION:

• Exceeds {{MAX_DOC_LINES}} lines
• Code mismatch
• Missing/inconsistent tags
• Excessive explanation

REJECTED:

• Document size uncontrollable (structural issue)
• Documentation itself impossible (code complexity issue)

---

test-reviewer (Step 8)

APPROVED:

• All test scenarios executed
• Works as intended
• Edge cases verified
• Performance criteria met

APPROVED_WITH_CONDITIONS:

• "Proceed after confirming {{X}} concurrent connections load test"
• "Proceed after production environment verification"

NEEDS_REVISION:

• Insufficient test coverage
• Performance criteria not met
• Test failures (code fix needed)

REJECTED:

• Testing itself impossible (architecture issue)
• Critical bugs (redesign needed)

---

Output Template

All reviewers must include this structure:

# [Agent Name] Review Summary

## Overall Assessment
[✅ APPROVED | ⚠️ APPROVED_WITH_CONDITIONS | 🔄 NEEDS_REVISION | 🚫 REJECTED]

[Additional sections based on selected status]

## [Status-Specific Required Sections]
[Follow output format for that status]

## Next Steps
[Actions conductor should take]

---

Conductor Processing Logic

<!-- CUSTOMIZE: Adapt to your tech stack -->

def handle_review_result(review_output):
    status = extract_status(review_output)

    if status == "APPROVED":
        proceed_to_next_step()

    elif status == "APPROVED_WITH_CONDITIONS":
        conditions = extract_conditions(review_output)
        verify_conditions(conditions)
        if all_conditions_met():
            proceed_to_next_step()
        else:
            return_to_previous_step()

    elif status == "NEEDS_REVISION":
        critical_issues = extract_critical_issues(review_output)
        if revision_count >= 3:
            escalate_to_user()  # Prevent infinite loop
        else:
            return_to_previous_step(issues=critical_issues)

    elif status == "REJECTED":
        stop_workflow()
        report_to_user(blocking_issues)
        wait_for_user_decision()

---

Consistency Checklist

Reviewer self-check before submitting review:

• Overall Assessment specifies one of 4 statuses?
• Using correct output format for selected status?
• Critical Issues truly critical? (NEEDS_REVISION/REJECTED only)
• Conditions verifiable by conductor? (APPROVED_WITH_CONDITIONS only)
• Next Steps clear?

---

Edge Case Decision Criteria

Ambiguous situations:

Situation	Choice
5 minor issues vs 1 critical issue	NEEDS_REVISION (critical takes priority)
90% complete, 10% missing	NEEDS_REVISION (100% completion required)
Cannot choose between alternatives A/B	REJECTED (user decision needed)
1 test fails, rest pass	NEEDS_REVISION (all tests must pass)
Document {{MAX_DOC_LINES + 10}} lines (standard: {{MAX_DOC_LINES}})	NEEDS_REVISION (clear criteria violation)

---

Remember: Consistency is key to automation. When all reviewers apply the same criteria, the conductor can manage workflows predictably.