一键导入
backlog-task-workflow
Operate Backlog.md tasks through the backlog CLI only; use when reading, planning, executing, or finalizing any task under backlog/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。
菜单
Operate Backlog.md tasks through the backlog CLI only; use when reading, planning, executing, or finalizing any task under backlog/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
Run and triage Python quality gates with minimal, root-cause fixes; use when validating feature work before completion.
Apply project testing standards for app/tests layout, naming, dependency overrides, and route/service coverage.
Apply typed FastAPI route patterns with clean dependency boundaries, stable error mapping, and test coverage for success/failure paths.
Author review-ready implementation plans grounded in real code, with a hard single-PR size gate and decomposition into safe incremental tasks when exceeded.
Apply partitioned settings and singleton provider patterns; use when adding or refactoring settings and provider wiring.
Apply pluggy registration and lifespan startup patterns for package discovery, initialization ordering, and testable startup behavior.
| name | backlog-task-workflow |
| description | Operate Backlog.md tasks through the backlog CLI only; use when reading, planning, executing, or finalizing any task under backlog/tasks. |
Use this skill whenever work is driven by a task in backlog/tasks/. The task file is the persistent record of scope, plan, and evidence; the CLI is the only supported way to mutate it.
backlog task edit / backlog task create. Never hand-edit task markdown: the SECTION:DESCRIPTION, AC:BEGIN, and DOD:BEGIN markers are CLI-owned and hand edits corrupt them.backlog task view <id> --plain (or backlog task list --plain). Plain output is the agent-friendly format.backlog instructions <guide> where guide is one of overview, task-creation, task-execution, task-finalization.auto_commit is false in backlog/config.yml: the CLI edits files but never commits. Git operations remain manual and user-controlled.| Intent | Command |
|---|---|
| Read a task | backlog task view TASK-1 --plain |
| Write/replace implementation plan | backlog task edit TASK-1 --plan "<multi-line text>" |
| Add an acceptance criterion | backlog task edit TASK-1 --ac "New criterion" |
| Check off AC / DoD item | backlog task edit TASK-1 --check-ac 2 --check-dod 1 |
| Start work | backlog task edit TASK-1 -s "In Progress" -a @me |
| Append a mid-work observation | backlog task edit TASK-1 --comment "Found X while doing Y" |
| Final implementation notes | backlog task edit TASK-1 --notes "<what changed, evidence>" |
| Create a follow-up/subtask | backlog task create "Title" -d "..." --ac "..." --dep TASK-1 --ref <path> |
Multi-line values: pass real newlines inside the quoted string (heredoc via "$(cat <<'EOF' ... EOF)" works well).
To Do → In Progress only when implementation actually starts, not during planning.--check-ac <index>) only when it is verified (its test passes), one by one, never batch-checked at the end.In Progress with notes written. Only a human moves a task to Done after verifying the Definition of Done.--comment and let the human decide (or use --remove-ac only with explicit approval).Work pauses for human review at three points; do not blow through them:
--plan is written, before any production code.--notes must state:
Done as an agent.