Description
Description
Currently, the agent task feature allows agents to spawn sub‑tasks without requiring any explicit permissions. In contrast, other operations like view, bash, write, and edit are properly gated by permission checks. This inconsistency enables agents to create a large number of unnecessary sub‑tasks, wasting time and resources.
Motivation
Our team’s “crush agent” frequently generates unauthorized sub‑tasks, many of which have no practical value. Because these sub‑tasks bypass the usual permission system, they can be easily abused or clutter the workflow. Aligning agent task with the rest of the permission model would prevent misuse and improve overall control.
Proposed Solution
Add a permission gate for agent sub‑tasks, consistent with the existing permission model for other agent actions.
Additional Context
Consistency across all agent capabilities will enhance security and user experience, mirroring how other actions already require explicit authorization.
Description
Description
Currently, the
agent taskfeature allows agents to spawn sub‑tasks without requiring any explicit permissions. In contrast, other operations likeview,bash,write, andeditare properly gated by permission checks. This inconsistency enables agents to create a large number of unnecessary sub‑tasks, wasting time and resources.Motivation
Our team’s “crush agent” frequently generates unauthorized sub‑tasks, many of which have no practical value. Because these sub‑tasks bypass the usual permission system, they can be easily abused or clutter the workflow. Aligning
agent taskwith the rest of the permission model would prevent misuse and improve overall control.Proposed Solution
Add a permission gate for agent sub‑tasks, consistent with the existing permission model for other agent actions.
Additional Context
Consistency across all agent capabilities will enhance security and user experience, mirroring how other actions already require explicit authorization.