Greenfield Planning Skill

Trigger

Run once per project: ws-skills plan

Reads BRIEF.md from the project root (or prompts for one). Asks 5 clarifying questions before planning.

Flow

Brief + Q&A
    ↓
Research agent (sequential)   ← domain landscape, ecosystem, hidden complexity
    ↓
┌─────────────────────────────────────┐
│ Architect │ PM │ Security (parallel)│  ← spawn 3 subagents simultaneously
└─────────────────────────────────────┘
    ↓
Synthesis             ← combines all into PLAN.md

Agents

1. Research (sequential — runs first)

Builds domain context that all other agents receive. Covers:

• Existing solutions and their weaknesses
• Standard and emerging tech stacks for this domain
• Technologies to avoid and why
• Non-obvious domain challenges
• Integration landscape (auth, payments, comms, etc.)
• Regulatory and compliance context
• Realistic timeline based on similar projects
• Hidden complexity — things that take 3x longer than expected
• Success patterns from the best products in this space

Output: research.md

2–4. Architect, PM, Security (parallel subagents)

Spawn all three simultaneously using the Agent tool. Each receives the full brief, Q&A, and research output as context.

Agent 1 — Architect
Agent 2 — PM           ← launch all three in a single message, do not wait
Agent 3 — Security

Wait for all three to complete before running Synthesis.

Architect output: architect.md

• System design overview
• Tech stack with rationale per layer
• Module boundaries and interfaces
• Infrastructure and CI/CD
• Scaling strategy
• Technical risks

PM output: pm.md

• Phases with tasks and acceptance criteria
• Success metrics
• Out of scope items
• Assumptions

Security output: security.md

• Threat model (likelihood + impact)
• Data classification
• Auth strategy recommendation
• Compliance flags
• Concrete security tasks
• Urgent flags

5. Synthesis

Combines research + all three persona outputs into a single PLAN.md covering:

• Overview
• Clarifications table
• Tech stack
• Architecture
• Phases with tasks and acceptance criteria
• Security plan
• Success metrics
• Risks
• Tensions (unresolved disagreements between personas)
• Assumptions
• Out of scope
• Branch naming (GIT-OS format)

Output: PLAN.md

Tools

Output Location

All files written to .wednesday/plans/ in the target directory:

.wednesday/plans/
├── research.md    ← domain context
├── architect.md   ← technical design
├── pm.md          ← phases and metrics
├── security.md    ← threat model
└── PLAN.md        ← combined PRD (primary output)

Failure Handling

Each agent fails independently. If one fails, the others continue and synthesis runs with whatever data is available. Failed agents show [partial fallback] in the progress display.

Rules

• Branch naming in PLAN.md must follow GIT-OS format
• Never generate CODEBASE.md for greenfield projects — it doesn't exist yet
• Cost target: under $0.20 per run