一键导入
plan
Use when creating implementation plans from specs or design documents - decomposes work into concrete TDD tasks with file paths and code examples
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when creating implementation plans from specs or design documents - decomposes work into concrete TDD tasks with file paths and code examples
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when ready to implement - selects a GitHub Issue, creates worktree, implements with TDD, verifies, reviews 5x, and creates PR. Numeric argument (e.g. pick-issue 3) is the issue number, not a parallelism count.
Use when you want to be exhaustively questioned about a plan or design until shared understanding is reached - asks one question at a time, traverses the design tree, and proposes a recommended answer for each question
Use to delegate all hands-on work to Codex - plan-writing, implementation, and refactoring. Opus reads, thinks, reviews, and writes docs only.
Use after a PR is opened and you want it merged - waits for CI to pass, marks draft PRs ready, removes the worktree, then merges and cleans up
Use when starting new feature work, adding functionality, or modifying behavior - explores design before implementation, creates plan, PR, and GitHub Issues
Use when encountering test failures, bugs, or unexpected behavior - requires root cause investigation before any fix attempt
| name | plan |
| description | Use when creating implementation plans from specs or design documents - decomposes work into concrete TDD tasks with file paths and code examples |
Creates bite-sized, actionable implementation plans from specs or design documents. Each task is a single TDD cycle (2-5 minutes).
Codex writes the plan; Opus reviews it. Codex writes plans more token-efficiently — broad, deep, and short — whereas Opus over-dives into detail when it writes a plan itself. So Opus stays the critic ("ツッコミ役"): delegate the plan-writing to Codex via the codex skill, then review what comes back against this skill's quality bar. Opus does not hand-write the plan; Opus does not edit the plan file directly — findings go back to Codex.
The steps below are the quality bar Opus reviews against, and the spec Opus hands to Codex — not a checklist for Opus to execute by hand.
Before writing tasks, map out:
Each file should have one clear responsibility.
Break work into tasks. Each task is one TDD cycle:
Each task includes:
Every task must have complete, concrete content. These are forbidden:
If you can't be specific, the design needs more work. Go back to brainstorming.
Use the correct test runner per language:
| Language | Test Command | Lint Command |
|---|---|---|
| Rust | cargo test | cargo clippy |
| Go | go test ./... | golangci-lint run |
| Ruby | bundle exec rspec or bundle exec rake test | bundle exec rubocop |
| TypeScript | npx vitest run or npx jest | npx eslint . |
| Terraform | terraform validate | terraform fmt -check |
Detect the project's actual setup and use the appropriate command.
Codex saves the plan to docs/plans/YYYY-MM-DD-<feature-name>.md in the
worktree. (A plan document is part of the docs Opus owns conceptually,
but the writing is still delegated to Codex — Opus directs and reviews.)
Opus reviews Codex's plan against the spec. Be the critic — poke holes, find gaps, name what's missing:
Findings go back to Codex to revise the plan. Do not rewrite the plan yourself — point out the problems and let Codex apply the fixes. Repeat review → revise until the plan clears the bar.