Delegation and Tooling
Helper agents carry out bounded tasks while you supervise the result. They can use available tools, report progress and return work to your notes, task board or project. Their permissions remain subject to the host and action policy.
A typical task
- Create a helper with a specific goal, such as reviewing a document.
- Check its status while it works.
- Inspect its result and attached evidence.
- Decide whether to incorporate the result or request further work.
Running, waiting, blocked, partial and completed are distinct states. A helper reporting completion is not an independent correctness check. A long answer is not proof of completion, and an unavailable tool is not a successful action.
delegate_subagent provides supported delegation through the normal tool flow.
The 3D workspace, Web and terminal report the shared task outcome where those
workflows are available.
Results and task boards
Supported helper results can remain available for inspection and project use. Check the reported save status before relying on recovery. Session history, current helper status and saved result artifacts are different records.
A projected task stays in Doing unless its supported task control or an
explicit project_column request changes the column. Review the result before
marking the task Done. “Ready for review” does not automatically certify
finished work.
A saved result does not authorize sending it. Preparing an email, document or other artifact still leaves any separate send or publication decision subject to the relevant approval rules.
Interruptions and failures
A blocked helper stops with a reported reason. Partial work can remain useful, but do not assume all requested steps completed. Cancellation cannot undo effects that already occurred.
If result storage fails, inspect the returned output and diagnostic. Do not repeat an external action automatically to repair a history problem. A saved conversation or imported bundle does not resume old approvals or provider work.
Coordinated workflows
Supported presets can divide a task among helpers, return their findings and stage follow-up actions. You remain responsible for reviewing important output and approving actions where the configured policy requires it. Audit or evidence records help review; they do not guarantee correctness or complete provenance.