elliottrjacobs

engineer-plan

Create structured implementation plans using parallel research agents. Use when the user says "plan this", "create a plan", "implementation plan", or after completing a tech spec. Breaks specifications into ordered, actionable tasks with effort estimates.

elliottrjacobs 1 Updated 6mo ago
GitHub

Install

npx skillscat add elliottrjacobs/bench-skills/engineer-plan

Install via the SkillsCat registry.

About this skill

This skill creates structured implementation plans by breaking technical specifications into ordered, actionable tasks with effort estimates. It solves the problem of translating high-level specs into concrete development steps by analyzing the codebase and researching best practices in parallel. Use it when a user requests a plan or after completing a tech spec, before beginning implementation work.

SKILL.md

/engineer-plan — Implementation Planning

Create structured implementation plans that break specs into ordered, actionable tasks. Plans bridge architecture (tech spec) and execution (engineer-work).

When to Use

  • User says "plan this", "create a plan", "implementation plan"
  • After completing a tech spec and ready to break it into tasks
  • When a feature needs structured decomposition before building

Process

Step 1: Gather Context

Check for existing specs (in priority order):

  1. docs/tech-specs/ — Most detailed, preferred source
  2. docs/prds/ — Product requirements if no tech spec
  3. User-provided description or $ARGUMENTS

Also check docs/solutions/ for relevant past solutions.

Step 2: Research (Parallel Agents)

Spawn 2 agents IN PARALLEL using the Task tool:

Agent 1 — Codebase Analyst

Task(subagent_type: "general-purpose", description: "Analyze codebase for planning")
prompt: Analyze this project to inform an implementation plan for [FEATURE].
  Focus on: existing patterns to follow, files/directories that need modification,
  related code that could be affected, test patterns, tech stack (check package.json,
  config files). Return findings relevant to planning.

Agent 2 — Practices Researcher

Task(subagent_type: "general-purpose", description: "Research best practices")
prompt: Research implementation best practices for [FEATURE TYPE] given the
  project's tech stack. Focus on: recommended implementation order, common
  integration challenges, testing strategies. Return practical guidance.

If the feature involves unfamiliar framework APIs or third-party integrations, launch an additional agent:

Agent 3 — Docs Researcher (conditional)

Task(subagent_type: "general-purpose", description: "Research framework docs")
prompt: Research [specific framework/API] documentation for [FEATURE].
  Check version compatibility with the project's package.json.
  Focus on: correct API usage, deprecation notices, common pitfalls.

Skip this agent if the feature uses well-established patterns already present in the codebase.

Step 3: Choose Detail Level

Use AskUserQuestion:

  • "What level of detail for this plan?" (header: "Detail")
    • "Minimal — task list with descriptions" — For small features, <1 day
    • "Standard — tasks with acceptance criteria (Recommended)" — For most features, 1 day - 1 week
    • "Comprehensive — detailed tasks with code guidance" — For complex features, >1 week

Step 4: Write Plan

Write the plan to docs/plans/YYYY-MM-DD-<name>.md.

Planning rules:

  • Tasks ordered by dependency (what must be done first)
  • Each task is independently implementable and testable
  • Include clear acceptance criteria (standard/comprehensive)
  • Mark parallel-safe tasks
  • Effort estimates: S (single file), M (multiple files), L (cross-cutting)
  • Include Task 0 for setup (branch, dependencies) if needed

Minimal plan format:

# Plan: [Feature Name]
**Date:** YYYY-MM-DD | **Spec:** [link] | **Detail:** Minimal

## Tasks
| # | Task | Description | Effort | Depends On |
|---|------|-------------|--------|------------|
| 0 | Setup | Branch, dependencies | S | — |
| 1 | ... | ... | S/M/L | — |

Standard plan format — adds per-task sections with:

  • Key files to modify
  • Acceptance criteria (2-4 bullets)
  • Test expectations
  • Dependency graph

Comprehensive plan format — adds per-task:

  • Implementation guidance with code patterns
  • Detailed test specifications
  • Rollback procedures
  • Risk assessment

Step 5: Handoff

Use AskUserQuestion:

  • "Plan is ready. What's next?" (header: "Next step")
    • "Start building (/engineer-work)" — Execute the plan
    • "Refine the plan" — Adjust tasks or detail
    • "Done for now" — Save and close

Output

Save to: docs/plans/YYYY-MM-DD-<name>.md

Next Steps

  • Ready to build? → /engineer-work
  • Need a tech spec first? → /product-tech-spec
  • Want to review after building? → /engineer-review