원클릭으로
writing-plans
Use when an approved local design spec exists and an exact implementation plan is needed before code changes
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Use when an approved local design spec exists and an exact implementation plan is needed before code changes
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | writing-plans |
| description | Use when an approved local design spec exists and an exact implementation plan is needed before code changes |
Write comprehensive implementation plans assuming the engineer has zero context for the codebase. Document exact files, interfaces, tests, commands, expected outputs, and commit boundaries. Use DRY, YAGNI, TDD, manual verification, and grouped commits.
Start only from an approved local spec path. Read the spec from disk and inspect the codebase fresh. Do not rely on prior conversation context.
If no spec path is provided, ask for it. Do not infer the feature from memory.
Read the approved spec's ### Phase Mode section. If the spec has no phase-mode section, read root AGENTS.md for a durable Superpowers Architecture phase-mode preference.
If neither the spec nor root AGENTS.md records a Phase Mode, ask the user which mode governs this workflow before writing the plan:
Should approval of this implementation plan start implementation in a fresh session automatically, or should implementation continue in this same session after approval?
Recommendation: Use the same mode selected during brainstorming. If no mode was selected, use automated fresh sessions for architecture-sensitive work because implementation starts from the approved plan with clean context.
Record the selected Phase Mode in the implementation plan header.
docs/superpowers/plans/YYYY-MM-DD-<feature-name>.md
Plans are local developer working state. Do not commit them. If docs/superpowers/** is not ignored, warn the user but do not edit .gitignore automatically.
If the approved spec covers multiple independent subsystems, suggest splitting it into separate implementation plans before writing a single oversized plan. Each plan must produce working, testable software on its own.
Before defining tasks, map out files to create, modify, inspect, or remove. Each file should have a clear responsibility. Follow existing codebase patterns over new abstractions.
Each step is one concrete action:
Task commit steps may commit code, tests, migrations, public docs, plugin metadata, and release assets.
Never include docs/superpowers/** in git add examples unless the user explicitly requests committing local Superpowers docs.
Every plan must start with:
# <Feature Name> Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Spec:** `<path-to-approved-spec>`
**Goal:** <one sentence>
**Phase Mode:** <selected phase mode and durability>
**Architecture:** <2-3 sentences using Design Understanding from the spec>
**Tech Stack:** <key technologies/libraries>
---
Do not write vague instructions such as empty marker tokens, future-work notes, broad error-handling requests, broad validation requests, test requests without exact test code, or cross-task references that require reading another task first.
If a step changes code, show the code. If a step validates behavior, show the exact command and expected result.
After writing the plan, ask one of these based on the selected Phase Mode.
Automated fresh-session mode:
Plan written to `<path>`. Please review it before implementation. After you approve it, I will start implementation in a fresh session using the selected automated fresh-session mode.
Same-session mode:
Plan written to `<path>`. Please review it before implementation. After you approve it, I will continue to implementation in this same session using the selected same-session mode.
If the user requests changes, update the plan and repeat the review gate.
After writing the plan, stop. Do not invoke implementation skills until the user explicitly approves the written plan.
After approval:
subagent-driven-development when tasks are mostly independent or executing-plans when they are linear, re-read the approved plan, referenced spec, and codebase from disk, and implement from the plan.Canonical implementation prompt:
Print the prompt that matches the approved plan shape:
subagent-driven-development for mostly independent tasks where fresh subagents can work task-by-task.executing-plans for tightly coupled, linear, or no-subagent execution.Prompt:
Use <implementation-skill> to implement:
<absolute-or-repo-relative-plan-path>
Read the plan, referenced spec, and codebase fresh. Use worktree isolation if I request it or if one is already active. Commit code per task, but never commit docs/superpowers/** unless I explicitly ask.
Before handing off the plan:
docs/superpowers/**.Use when starting a new project from scratch or establishing project operating contracts, root CONTEXT.md, stack decisions, architecture summary, domain language, and a high-level roadmap before feature brainstorming
Use when executing implementation plans with mostly independent tasks in a fresh or same session after written plan approval
Use when starting a conversation to route new-project setup requests to project-setup and make/build/create/implement app, site, tool, component, UI, API, workflow, or behavior requests to brainstorming before frontend, design, framework, or implementation skills
Use before make/build/create/implement requests for apps, sites, tools, components, UI, APIs, workflows, feature work, architecture changes, behavior changes, or unclear requirements; writes a local design spec before planning or coding
Use when executing written implementation plans linearly after written plan approval
Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.