Three-agent pipeline (Planner → Architect → Critic) that produces a consensus plan before implementation. Requires explicit human approval at the gate before proceeding to /dog:prp-implement. Use when a feature or refactor is complex enough that a single agent's plan could miss blind spots.

/dog:consensus-planning — Consensus Planning Pipeline

Three-agent plan review cycle that surfaces blind spots, disagreements, and edge cases before any code is written. Stops at a human-approval gate.

When to Use

Features touching more than 3 modules
Architecture decisions with long-term implications
Any task where the wrong plan would cause painful rework
When /dog:prp-plan output feels uncertain or ambiguous

When NOT to Use

Quick bug fixes — use /dog:plan instead
Tasks already fully specified with no open questions — jump to /dog:prp-implement
Single-file changes — overhead not justified

Pipeline

[Planner agent] → initial proposal
      ↓
[Architect agent] → viability + risk review
      ↓
[Critic prompt] → blind spots + edge cases
      ↓
[Human gate] → explicit approval required
      ↓
/dog:prp-implement

Process

Step 1 — Planner produces initial proposal

Invoke the planner agent with the task description:

"Produce a structured plan for: {task}. Include: goals, approach, file list, risks, open questions."

Save output as .dog/PRPs/plans/consensus-{slug}.plan.md (draft).

Step 2 — Architect reviews viability + risks

Invoke the architect agent with the planner's output:

"Review this plan for technical viability and risks. Flag: (1) architectural concerns, (2) missing dependencies, (3) integration risks, (4) scalability issues. Be specific — line-level feedback preferred."

Append architect feedback to the plan file under ## Architect Review.

Step 3 — Critic surfaces blind spots

Run a focused critic prompt (inline, no separate agent):

"You are a skeptical senior engineer. Read this plan and find: (1) assumptions that could be wrong, (2) edge cases not covered, (3) anything that will break in production that the plan doesn't address. List each as a numbered item with severity: HIGH / MEDIUM / LOW."

Append under ## Critic Review.

Step 4 — Human approval gate

Present the complete plan (proposal + architect review + critic review) to the user and ask:

"Consensus plan ready for review.

Three agents have reviewed this plan. Do you approve proceeding to implementation?

Yes → run /dog:prp-implement .dog/PRPs/plans/consensus-{slug}.plan.md

No → describe what to change and I'll revise"

Do not proceed to implementation without explicit approval.

Step 5 — Revise if needed

If the user requests changes: update the plan and repeat Steps 2-4.

If approved: hand off to /dog:prp-implement.

Verification

A consensus planning session is complete when:

The plan file exists at .dog/PRPs/plans/consensus-*.plan.md
It contains all three sections: proposal, architect review, critic review
The user has explicitly approved (in writing in the conversation)

Skill: dog:writing-plans, dog:blueprint
Command: /dog:prp-plan, /dog:prp-implement
Agent: planner, architect

consensus-planning