一键导入
finishing-task
Use only when the user explicitly asks to finish a tracked `.my-context/tasks/` record and it needs durable wrap-up and memory promotion.
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 only when the user explicitly asks to finish a tracked `.my-context/tasks/` record and it needs durable wrap-up and memory promotion.
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
Helps explore ideas before implementation. Use when the user wants to brainstorm features, compare approaches, shape requirements, or discuss tradeoffs before writing code.
Use when you have two or more independent tasks, failures, or investigations that can be worked on concurrently without shared state or sequential dependencies.
Use when receiving code review feedback, before implementing suggestions, especially when feedback is unclear, broad, or may not fit the codebase as-is.
Use when finishing substantial code changes, before merging, or at natural checkpoints to request an explicit code review pass.
Use when executing a multi-task implementation plan by dispatching one fresh builder agent per task, reviewing each task before moving on, and keeping controller context narrow.
Use when investigating a bug, test failure, build failure, flaky behavior, or unexpected runtime issue before proposing or implementing fixes.
| name | finishing-task |
| description | Use only when the user explicitly asks to finish a tracked `.my-context/tasks/` record and it needs durable wrap-up and memory promotion. |
Use this skill only when the user explicitly asks to close, finish, or complete a tracked task. It defines the completion quality gate; task_finish performs the task completion operation.
Do not invoke this skill automatically at the end of implementation, review, or a session.
Before proposing or performing completion, confirm as applicable that:
Do not mark a task complete because coding started, implementation seems done, or the user paused. Finishing the task record does not create a PR or close an external ticket; those remain separate human actions and require explicit user instruction.
The durable record must reflect what actually happened. Before finishing, explicitly check:
context-memoryRequired durable documentation and durable memory updates are part of finishing, not optional follow-up. Search existing .my-context/ documentation first; update the best existing document where possible, and create one only when no existing document fits.
Prepare a concise factual outcome for task_finish:
Do not include session chatter or describe merely planned work as delivered.
After the user explicitly directs completion and all gates above pass:
task_finish with the factual outcome.context-memory.task_finish writes the outcome and performs the completed-state move. Do not move task folders or edit state.json directly.
Update existing durable documentation when a completed task changes documented behavior. Add a durable document only when a new component, flow, or long-lived concept lacks a suitable home. Do not create parallel documentation when an existing document should be updated.
Do not:
task_finish without explicit user instructionoutcome.md, task folders, or state.json directlycontext-memoryOn explicit user direction, verify the work and durable record, prepare a factual outcome, finish with task_finish, and preserve only future-useful learnings in durable memory.