Architect: Request Review

Architect: Request Review

Send PROPOSAL-N task to reviewers for feedback.

-> Full workflow in PROCESS.md

When to Use

Plan draft complete, ready for review.

Given/When/Then/Should

Given plan ready when requesting review then spawn 3 axis-specific reviewers (A=Correctness, B=Test quality, C=Elegance) should never spawn reviewers without axis assignment

Given reviewers when assigning then provide Beads task ID and context should never expect reviewers to search

REVIEW Naming

Reviews are named PROPOSAL-N-REVIEW-{axis}-{round} where:

• N = proposal number (matches PROPOSAL-N)
• axis = reviewer criteria axis (A, B, or C)
• round = review round number (1, 2, ...)

Review Axes

Axis	Focus	Key Questions
A	Correctness (spirit and technicality)	Does it faithfully serve the user? Are technical decisions consistent with rationale?
B	Test quality	Integration over unit? SUT not mocked? Shared fixtures? Assert outcomes?
C	Elegance and complexity matching	Right API? Not over/under-engineered? Complexity proportional to problem?

Steps

1. Verify PROPOSAL-N task is complete with all sections
2. Spawn three reviewers with the task ID and URD reference:

Task(description: "Reviewer A: correctness", prompt: "Review PROPOSAL-1 task <task-id>. URD: <urd-id> (read for requirements context). You are Reviewer A (Correctness). Focus: Does it faithfully serve the user? Are technical decisions consistent with rationale? Create review task titled PROPOSAL-1-REVIEW-A-1...", subagent_type: "reviewer")
Task(description: "Reviewer B: test quality", prompt: "Review PROPOSAL-1 task <task-id>. URD: <urd-id> (read for requirements context). You are Reviewer B (Test quality). Focus: Integration over unit? SUT not mocked? Shared fixtures? Assert outcomes? Create review task titled PROPOSAL-1-REVIEW-B-1...", subagent_type: "reviewer")
Task(description: "Reviewer C: elegance", prompt: "Review PROPOSAL-1 task <task-id>. URD: <urd-id> (read for requirements context). You are Reviewer C (Elegance). Focus: Right API? Not over/under-engineered? Complexity proportional to problem? Create review task titled PROPOSAL-1-REVIEW-C-1...", subagent_type: "reviewer")

3. Wait for all 3 reviewers to vote ACCEPT

Consensus

All 3 reviewers must vote ACCEPT. Max revision rounds until consensus.

Checking Reviews

bd show <proposal-id>
bd comments <proposal-id>

Coordination

# Add comment to notify that review is ready
bd comments add <proposal-id> "Review requested — 3 reviewers spawned"

# Check for review votes
bd comments <proposal-id>

Follow-up Proposal Reviews (FOLLOWUP_PROPOSAL-N)

For FOLLOWUP_PROPOSAL-N reviews, use the same procedure:

• Review task naming: FOLLOWUP_PROPOSAL-N-REVIEW-{axis}-{round}
• Same 3 axes (A/B/C), same binary ACCEPT/REVISE vote
• No severity tree for plan reviews (same as original plan reviews)
• Reviewers should also verify that FOLLOWUP_PROPOSAL addresses the specific IMPORTANT/MINOR findings scoped in FOLLOWUP_URE/URD