1.8 KiB
1.8 KiB
argument-hint, description
| argument-hint | description |
|---|---|
| optional instructions | Process all tasks automatically |
Process all tasks automatically.
Repeatedly work through incomplete tasks from the project task list.
If the user provided additional instructions, they will appear here:
$ARGUMENTSIf the user did not provide instructions, work through ALL incomplete tasks until NONE remain.
Steps
- Track attempt count and previously attempted tasks to prevent infinite loops
- Use the
@tasksskill to extract the first incomplete task from.llm/todo.md - If a task is found:
- Check if we have already attempted this task 1 time
- If yes, mark it as blocked (with
- [!]) and continue to next task - If no, launch the
@tasks:do-taskagent to implement it - Do NOT add instructions to the agent prompt - the agent is self-contained and follows its own workflow (including precommit, commit, rebase)
- Do NOT mark the task as complete yourself - the
do-taskagent does this
- Repeat until no incomplete tasks remain or the user's instructions are met
- When all tasks are completed, archive the task list:
@../shared/scripts/task-archive.md
Notes
- Each task is handled completely by the
do-taskagent before moving to the next - The
do-taskagent marks tasks as complete - do NOT calltask_complete.pyyourself - Each task gets its own commit for clear history
- After each agent returns, check the task list again to see if more tasks remain
User feedback
Throughout the process, provide clear status updates:
- "Starting task: [task description]"
- "Task completed successfully: [task description]"
- "Task failed: [task description]"
- "Skipping blocked task: [task description]"
- "All tasks completed - task list archived to .llm/YYYY-MM-DD-todo.md" or "Stopping due to failures"