メニュー
Repeatedly check if there is more work to do until the task is truly complete. Use after implementation to ensure nothing was missed against the acceptance criteria.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use task-factory CLI commands to manage workspaces, tasks, and workflows. Use when the user needs to create tasks, manage task execution, check queue status, or work with the Task Factory system via CLI.
Capture a browser screenshot for task validation and attach it to the active task using attach_task_file.
Create a branch, commit all agent changes, push, and open a pull request. Use when work is ready for review and should be submitted as a PR rather than just checkpointed.
Commit and push the agent's current work without opening a PR. Stages only files modified by the agent, creates a descriptive commit message, and pushes to the current branch.
Review all code changes for bugs, edge cases, naming, complexity, and missing error handling. Use after completing implementation work to catch issues before marking a task complete.
Scan code changes for security vulnerabilities including OWASP Top 10, hardcoded secrets, injection flaws, and authentication issues. Use after implementation to catch security problems before marking a task complete.
| name | grind-loop |
| description | Repeatedly check if there is more work to do until the task is truly complete. Use after implementation to ensure nothing was missed against the acceptance criteria. |
| metadata | {"author":"task-factory","version":"1.0","type":"loop","hooks":"pre-planning,pre,post","max-iterations":"5","done-signal":"HOOK_DONE"} |
| config | [{"key":"max-iterations","label":"Max Iterations","type":"number","default":"5","description":"Maximum number of grind loop iterations","validation":{"min":1,"max":20}}] |
Review everything you have done for this task. Check your work against the acceptance criteria and testing instructions.
If everything is truly complete and correct, respond with exactly:
HOOK_DONE
Otherwise, fix what is needed and continue working. Do not respond with HOOK_DONE until you are confident the task is fully complete.
