Phase 4: Design Reflection - Detailed Guidance

Goal

Use parallel sub-agents to systematically review design documents and identify gaps before committing.

Why Reflection Matters

Design documents can have issues that impact implementation:

• Requirements from Phase 1 that got lost in synthesis
• Missing BDD scenarios for edge cases or error conditions
• Inconsistencies between documents
• Undocumented assumptions and risks

Reflection catches these issues before implementation begins.

Why Use Sub-Agents

Sub-agents provide:

• Fresh perspective - No bias from having written the documents
• Parallel execution - Multiple review aspects simultaneously
• Specialized focus - Each agent focuses on one aspect deeply
• Objective analysis - Independent assessment without context attachment

Sub-Agent Launch Strategy

Core Reflection Sub-Agents (Always Required)

Launch these three sub-agents in parallel using the Agent tool with subagent_type=general-purpose:

Sub-agent 1: Requirements Traceability Review

You are reviewing design documents for requirements coverage.

Context: [Provide Phase 1 requirements summary]

Your task:
1. Read all design documents in docs/plans/YYYY-MM-DD-<topic>-design/
2. Create a traceability matrix mapping each Phase 1 requirement to where it's addressed
3. Identify any orphaned requirements (not addressed anywhere)
4. Identify any implementation details without corresponding requirements

Output format:
- Requirements Traceability Matrix (requirement → document → section)
- Orphaned Requirements List (requirements not addressed)
- Scope Creep List (implementation without requirements)

Sub-agent 2: BDD Completeness Review

You are reviewing BDD specifications for completeness.

Context: [Provide feature summary]

Your task:
1. Read bdd-specs.md in docs/plans/YYYY-MM-DD-<topic>-design/
2. Categorize all scenarios as: happy path, error path, or edge case
3. Identify missing scenarios for each category
4. Check that each scenario has complete Given-When-Then structure

Output format:
- Scenario Coverage Summary (count by category)
- Missing Happy Path Scenarios
- Missing Error Path Scenarios (validation, auth, external failures, timeouts)
- Missing Edge Case Scenarios (boundaries, empty states, concurrency)
- Incomplete Scenarios (missing Gherkin structure)

Sub-agent 3: Cross-Document Consistency Review

You are reviewing design documents for consistency.

Your task:
1. Read all design documents in docs/plans/YYYY-MM-DD-<topic>-design/
2. Build a terminology glossary from all documents
3. Identify terminology inconsistencies (same concept, different terms)
4. Verify cross-references between documents work
5. Check component/file names are consistent across documents

Output format:
- Terminology Glossary (term → definition → source document)
- Inconsistencies Found (term variations, conflicting definitions)
- Broken Cross-References (links that don't resolve)
- Naming Inconsistencies (component/file name variations)

Additional Reflection Sub-Agents (Launch as Needed)

Security Review Sub-agent (for features with security implications):

You are reviewing design documents for security considerations.

Context: [Provide feature summary and threat context]

Your task:
1. Read all design documents in docs/plans/YYYY-MM-DD-<topic>-design/
2. Identify security-relevant components and data flows
3. Check for documented threat model and mitigations
4. Identify potential vulnerabilities not addressed

Output format:
- Security-Relevant Components
- Threat Model Coverage (what's addressed vs missing)
- Unaddressed Security Concerns
- Recommendations

Risk Assessment Sub-agent (for complex or high-stakes features):

You are reviewing design documents for risks and assumptions.

Context: [Provide feature summary and constraints]

Your task:
1. Read all design documents in docs/plans/YYYY-MM-DD-<topic>-design/
2. List all explicit assumptions found
3. Identify implicit assumptions (things assumed but not stated)
4. Identify technical, integration, and implementation risks
5. Check if risks have documented mitigations

Output format:
- Explicit Assumptions List
- Implicit Assumptions List (things taken for granted)
- Technical Risks (complexity, performance, scalability)
- Integration Risks (dependencies, APIs, data migration)
- Implementation Risks (new code, changes to existing code, infrastructure)
- Risks Without Mitigations

Integration Workflow

1. Collect Sub-Agent Results

Use TaskOutput tool to retrieve results from all launched sub-agents.

2. Synthesize Findings

Merge findings into a unified gap list:

Category	Finding	Priority	Document to Update
Orphaned requirement	"X not addressed"	High	_index.md
Missing scenario	"Error case Y"	High	bdd-specs.md
Inconsistency	"Term Z varies"	Medium	All
Unaddressed risk	"Dependency W"	Medium	best-practices.md

3. Prioritize Gaps

High Priority (must fix before commit):

• Orphaned requirements
• Missing happy path scenarios
• Security vulnerabilities
• Critical risks without mitigation

Medium Priority (should fix):

• Missing error/edge case scenarios
• Terminology inconsistencies
• Broken cross-references
• Implicit assumptions

Low Priority (nice to have):

• Additional documentation
• Diagrams
• Examples

4. Update Documents

Based on prioritized gap list:

1. Fill gaps - Add missing sections, scenarios, or details
2. Clarify ambiguities - Make implicit things explicit
3. Document assumptions - State what was assumed
4. Add risk mitigations - Address identified risks
5. Fix inconsistencies - Align terminology and references

5. Re-Verify Updated Sections

For significant updates, consider launching a quick verification sub-agent:

You are verifying that specific gaps have been addressed.

Gaps that were identified:
[List the specific gaps that were fixed]

Your task:
1. Read the updated sections in docs/plans/YYYY-MM-DD-<topic>-design/
2. Verify each gap is now addressed
3. Report any gaps that remain

Output format:
- Verification Results (gap → addressed: yes/no)
- Remaining Issues

Output

• Updated design documents with gaps filled
• All Phase 1 requirements traced to implementation
• Complete BDD scenario coverage
• Consistent terminology across documents
• Documented assumptions and mitigations

Anti-Patterns to Avoid

• Skipping reflection - "It looks fine, let's just commit"
• Single-agent reflection - Not leveraging parallel sub-agents
• Ignoring sub-agent findings - Not acting on identified gaps
• Superficial fixes - Adding content without proper integration
• Inconsistent fixes - Fixing in one document but not others