do-all-tasks

Process all tasks automatically using Claude Code's built-in task tools.

Repeatedly work through incomplete tasks from the built-in task list.

If the user provided additional instructions, they will appear here:

If the user did not provide instructions, work through ALL incomplete tasks until NONE remain.

Steps

1. Track attempt count and previously attempted tasks to prevent infinite loops
2. Call TaskList to find the first pending, unblocked task
3. If a task is found:
   • Check if we have already attempted this task 1 time
   • If yes, skip it and continue to the next task
   • If no, launch the builtin-tasks:do-task agent 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-task agent does this
4. Verify HEAD tests good before the next task. After the do-task agent returns, run git test run HEAD --retest --verbose --verbose (cached, so near-instant on an already-passing commit). If it does not return good (non-zero exit, or git test results HEAD reports bad/unknown), stop the loop and report — do not stack the next task on a broken base.
5. Repeat until no incomplete tasks remain or the user's instructions are met
6. When all tasks are completed, report the final status

Notes

• Each task is handled completely by the do-task agent before moving to the next
• The do-task agent marks tasks as complete — do NOT call TaskUpdate yourself
• Each task gets its own commit for clear history
• After each agent returns, call TaskList again to see if more tasks remain

User feedback

Throughout the process, provide clear status updates:

• "Starting task: [task subject]"
• "Task completed successfully: [task subject]"
• "Task failed: [task subject]"
• "Skipping previously failed task: [task subject]"
• "All tasks completed" or "Stopping due to failures"