一键导入
task
Create a new task (bug, refactoring, or qa) with spec and progress tracking. Use when user wants to create a task or mentions /task command.
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
Create a new task (bug, refactoring, or qa) with spec and progress tracking. Use when user wants to create a task or mentions /task command.
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
Review the latest work unit with fresh, unnarrowed eyes and surface any problem it contains as follow-ups. Use after a work-unit commit, or when the user runs /agent-review.
Continue working on a story or task by reading progress.md, executing the next work unit, and updating progress. Use when user wants to resume story/task work or mentions /continue command.
Run a Selenium test in visible (non-headless) mode with slowdown so the user can watch it. Use when user wants to demo or visually watch a Selenium test or mentions /demo command.
Preview the planned design for a scenario before writing tests. Shows domain model changes, usecase patterns, and key design choices. Use before red-usecase to get user approval on the implementation approach. If the user rejects the design, offers to discuss inline or escalate to /architecture for a full ADR.
Write a journey summary capturing noteworthy moments (predictions, decisions, surprises, mistakes, quirks). Runs automatically the moment the agent observes something worth noting during work, and manually before /clear or /compact as a final sweep. Idempotent — skips entries already recorded. Use when the agent observes a noteworthy moment, before clearing context, or when the user mentions /handoff.
Re-explain the last thing in plain words. Use ONLY when the user explicitly types /plain or asks to "explain that simply / in plain words / like I'm not an expert". Never invoke on your own initiative — this is a manual button the user presses when something was hard to follow.
| name | task |
| description | Create a new task (bug, refactoring, or qa) with spec and progress tracking. Use when user wants to create a task or mentions /task command. |
bug, refactoring, or qaList ALL directories under ProductSpecification/tasks/ AND ProductSpecification/tasks/done/ (use ls). For EACH folder, extract the leading integer (e.g., 1-refactoring-task-entity -> 1, 2-bug-modal-scroll -> 2). List every extracted number explicitly, then pick the max. Next number = max + 1. If no folders exist, start at 1. All task types share one global sequence -- a bug after refactoring task 1 gets number 2, not 1. Done tasks still occupy their numbers.
Verification (mandatory): Before creating the folder, confirm that NO existing folder starts with the chosen number. Run ls ProductSpecification/tasks/ | grep "^{N}-" -- if it returns results, increment N and re-check.
If arguments provided, parse type (bug, refactoring, or qa) and title from args.
If no arguments, ask interactively: type and short title (2-5 words).
Create ProductSpecification/tasks/{N}-{type}-{slug}/ where slug is lowercase-hyphenated title.
Gather from user:
Bug: Problem (as thorough as possible — symptoms, observed vs. expected, environment, frequency, any captured response/error) and Reproduction steps. Do NOT gather a Solution, Affected Layers, or Key Files at creation — those are produced by the discovery sequence (root cause analysis records the cause and key files in spec.md; design settles the fix approach; steps discovery scopes the layers). Describe the problem fully; defer every claim about the fix.
Refactoring: Problem, Solution, Affected Layers (domain, usecase, h2, rest, email, frontend), Key Files, and numbered steps with clear scope.
QA only: Problem (why this checklist exists), Solution (when to run, environment, session duration), Cases (numbered one-line items expressing intent — no Gherkin, no implementation detail). No Affected Layers, no Key Files (QA tasks don't change code).
Write ProductSpecification/tasks/{N}-{type}-{slug}/spec.md using the format in .claude/templates/task/creation-formats.md.
Select fix profile based on type and affected layers:
| Type / Affected Layers | Section |
|---|---|
| Bug — any layer (backend, frontend, or both) | ## Fix: {description} discovery-first: [ ] root cause analysis → [ ] design → [ ] steps discovery. Prepend [ ] reproduce in prod-copy when the bug is observed in prod-copy. Concrete TDD steps (including any adapters-discovery) are inserted by /continue at steps discovery. |
| Refactoring | ## Fix with user-defined steps |
| QA | ## Cases with one [ ] checkbox per case (no TDD sub-steps) |
Write ProductSpecification/tasks/{N}-{type}-{slug}/progress.md using the matching format from the template. Bug tasks are discovery-first regardless of affected layer — never pre-plant adapters-discovery at creation; it is a sub-step steps discovery inserts, not the bug gate. Refactoring steps are user-defined from step 4. QA tasks mirror the ## Cases section of spec.md as checkboxes — the tester ticks them during a session.
Show spec.md and progress.md to user for review. Commit both files.
task: spec (Task {N}, {title}).claude/templates/task/creation-formats.md -- spec.md format, progress.md formats