onex-writing-plans

name	onex-writing-plans
description	Use this skill when the user has a completed design doc and needs a comprehensive implementation plan with exact file paths, complete code examples, and a TDD approach.
disable-model-invocation	true

onex-writing-plans

Use this skill when the user has a completed design doc and needs a comprehensive implementation plan with exact file paths, complete code examples, and a TDD approach.

Purpose

Convert a validated design into a step-by-step implementation plan that any engineer can follow without prior codebase context. Each task is one action, 2-5 minutes of work.

Prerequisites

• A completed design document (from brainstorming or provided inline)
• The design doc path (e.g., docs/plans/YYYY-MM-DD-<topic>-design.md) or pasted content

Workflow

1. Read the design document. If a path is provided, read the design doc. If no design is provided, ask the user to provide it before proceeding.

2. Write the plan header. Every plan starts with: feature name, goal (one sentence), architecture (2-3 sentences), and tech stack.

3. Break work into phases. Each phase covers one component. Within each phase, tasks follow strict TDD order:

   • Write the failing test (complete code, not a placeholder)
   • Run the test to confirm it fails (exact command + expected error)
   • Write minimal implementation (complete code)
   • Run the test to confirm it passes (exact command)
   • Commit (exact git command)

4. Ensure bite-sized granularity. Each step is one action, 2-5 minutes. Never write "add validation logic" as a single step. Break into: write test, run, implement, verify, commit.

5. Run adversarial review (R1-R6). Before presenting the final plan, run these checks:

   • R1 — Count integrity: numeric quantifiers match actual list items
   • R2 — Acceptance criteria strength: no vague language, all measurable
   • R3 — Scope violations: each task can actually deliver what it claims
   • R4 — Integration traps: import paths, contract paths, API signatures verified
   • R5 — Idempotency: resource creation has dedup keys and rerun behavior
   • R6 — Verification soundness: each verification step graded strong/medium/weak

6. Save the plan and hand off. Save to docs/plans/YYYY-MM-DD-<feature-name>.md. Output the handoff line referencing the path and pointing to 12-plan-ticket.

Expected Output Format

A Markdown plan document with:

• Header (goal, architecture, tech stack)
• Phases with exact file paths (Create/Modify/Test)
• Complete code for each step (not pseudocode)
• Exact run commands with expected output
• Adversarial review summary (R1-R6)
• Handoff line with artifact path

Quality Checklist

• Plan header includes goal, architecture, and tech stack
• Every step has exact file paths — no "somewhere in src/"
• Complete code provided — no "add validation here" placeholders
• TDD order followed: test first, then implementation
• Each step is one action, 2-5 minutes of work
• Adversarial review (R1-R6) was run and results documented
• Plan saved to docs/plans/YYYY-MM-DD-<feature-name>.md
• Handoff line references the actual saved artifact path