planner

Product and engineering planning agent with two-lens review that produces structured plans and enforces Gate 1.

Overview

The planner creates structured implementation plans before any code is written. It applies two lenses: a product lens (is this the right thing to build?) and an engineering lens (is this the right way to build it?). The output is a plan file at tasks/plans/YYMMDD-name.md that must be approved by a human before implementation begins (Gate 1).

The planner challenges your assumptions, evaluates alternatives, estimates effort, and flags risks — all before a single line of code is touched.

Quick Reference

Plan File Structure

Every plan MUST include these six sections:

Section	Purpose
Problem Statement	What problem this solves and for whom
Success Criteria	Measurable definition of done
Out of Scope	What this plan explicitly does NOT cover
Technical Approach	How it will be built (architecture, components, data flow)
Risk Flags	What could go wrong and mitigation strategies
Estimated Effort	Time estimate for both human team and MeowKit

Two-Lens Evaluation

Lens	Questions asked
Product	Is this the right thing to build? Are assumptions validated? Does the user really need this?
Engineering	Is the approach sound? Are there simpler alternatives? What are the tradeoffs?

How to Use

# Explicit invocation
/meow:plan add user authentication with JWT

# As part of the cook pipeline
/meow:cook add shopping cart  # → planner runs first automatically

The planner will:

1. Challenge your premises ("Do you really need JWT? Have you considered session-based auth?")
2. Apply both lenses
3. Produce a plan file
4. Wait for your approval (Gate 1)

Under the Hood

Handoff Example

Planner receives from orchestrator:

Task: "Add user authentication with JWT"
Complexity: COMPLEX
Sequence position: Phase 1

Planner produces:

File: tasks/plans/260327-auth-jwt.md
Status: AWAITING APPROVAL (Gate 1)
Handoff to: architect (if architectural decisions needed), then tester (Phase 2)
Risk flags: 2 (session management complexity, token refresh edge cases)

After human approves: Pipeline continues to tester (or architect first if flagged).

Troubleshooting

Issue	Cause	Fix
Plan rejected by user	Scope too broad or wrong approach	Ask planner to narrow scope or explore alternatives
Planner asks too many questions	Task is genuinely ambiguous	Provide more context: target users, constraints, existing patterns
Plan missing sections	Shouldn't happen (enforced)	Report — all 6 sections are mandatory
Can't proceed past Gate 1	Plan not yet approved	Review and type "approve" to continue