菜单
Decompose a user story into ordered implementation tasks with codebase references. Use when breaking down a story, planning tasks, or preparing work for implementation. Triggers on: decompose story, break down story, plan tasks, task breakdown, decompose to tasks.
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
Implement a feature fully autonomously in the background as a Workflow — decompose, then run each task through test-design → implement → refactor → review → verify, closing on executed evidence instead of human approval gates. Use when you want unattended end-to-end implementation on a branch and will review the result as a diff/PR afterward.
Deliver a large user story as multiple parallel PRs. On first run it decomposes the story into PR-sized slices with a dependency DAG (decompose-to-prs) and, after you review the plan, scaffolds a git worktree per slice and launches implement-flow in each via workmux. Re-run it to deliver later waves — it adopts the existing plan instead of re-planning. Use when a story is too big for one PR.
Implement a feature autonomously through the full test-design → test-write → implement → refactor → review loop, pausing only for plan approval and pre-commit approval.
Implement a feature autonomously through the full test-design → test-write → implement → refactor → review loop, running independent tasks in parallel (isolated git worktrees) and pausing only for plan approval and pre-commit approval.
Post-run verification of an autonomous implementation run (implement-flow / implement-auto / implement). Runs the run-verifier agent to catch the failure modes the evidence gate can't see — staged tails, dead code, vacuous receipts, collapsed commits — and, with --fix, remediates them. Use after an unattended run, before trusting its "done".
Write or edit an article or note. Follows writing style guidelines for clarity and consistency.
| name | decompose-to-tasks |
| description | Decompose a user story into ordered implementation tasks with codebase references. Use when breaking down a story, planning tasks, or preparing work for implementation. Triggers on: decompose story, break down story, plan tasks, task breakdown, decompose to tasks. |
| user-invocable | true |
| argument-hint | <user-story-file-or-description> |
Break a user story into an ordered list of implementation tasks with codebase references, ready for /implement or /tdd.
Important: Do NOT start implementing. Just create the task list.
Accept the user story from $ARGUMENTS — this can be a file path, inline text, or pasted block.
Ask only if the input is ambiguous or lacks enough detail to decompose. If the user story is already well-specified (has clear acceptance criteria), skip or limit to 1-2 questions.
Focus on:
Format questions for quick response:
1. Which acceptance criteria should be tackled first?
A. Start with the core happy path (US-001, US-002)
B. Start with the riskiest unknown (US-003)
C. Follow the dependency order as written
D. Other: [please specify]
2. Are there any existing patterns in the codebase we should align with?
A. Yes — I'll point you to the relevant code
B. No — this is a new area
C. Not sure — please explore and recommend
D. Other: [please specify]
This lets users respond with "1A, 2C" for quick iteration. Indent the options.
Delegate the core decomposition work to the decompose-to-tasks agent using the Task tool:
Decompose the following user story into implementation tasks:
[user story content here]
[Include any clarified scope, ordering preferences, or constraints from the user's answers]
The agent will:
tasks/[story-name].mdShow the user the task summary returned by the agent, including:
After presenting the task list, ask the user:
If the user requests changes, delegate back to the agent with the specific feedback to update the file in place.