[Feature]: Scheduled follow-ups for issue monitoring and permission-scoped replies

[travelingflwr]

Request

Add a built-in scheduler for agent follow-ups in an existing T3 Code conversation, including monitoring external work and responding within user-defined permissions.

Concrete use case

I used T3 Code to file a GitHub bug report. I then wanted the agent to periodically check that issue for maintainer replies, tell me when something needs attention, and help respond using the context already collected in the conversation.

The missing piece in this workflow is a way to schedule those follow-ups from T3 Code itself, rather than moving the task and its context into another application or building an external cron job.

Suggested behavior

• Schedule a saved follow-up instruction from an existing chat, with one-time or recurring timing, timezone, and optional active hours.
• Keep the original conversation context and link each scheduled run back to that thread.
• Offer pause/resume, edit, run now, delete, next-run status, and a readable run history.
• Support change-only notifications: do not produce repeated alerts or comments when the watched issue has not changed.
• Let users choose read-and-notify, draft-for-approval, or explicitly authorized automatic replies. Default external writes to approval; scheduling a check should not itself authorize posting.
• For automatic replies, constrain the agent to confirmed facts and the user's defined scope. Never claim a fix was tested without a real test, or treat a maintainer comment as permission to install software or change the local machine.
• Prevent overlapping runs and duplicate replies. Surface expired authentication, unavailable tools, and failed checks clearly.
• Provide an end date or stop condition, such as notifying on issue closure and pausing the schedule.
• Make execution requirements explicit: which host/server runs the task, whether it must remain online, and what happens to missed runs. Closing a client should not silently imply background execution is guaranteed.

Initial scope

A time-based scheduler attached to one existing thread would be enough; this does not require a hosted service or GitHub webhooks in the first version. Ideally, schedule state and results would be visible across desktop, web, and mobile clients connected to the same server, with execution permissions enforced on the server.

Related proposals

• Recurring autonomous runs in one T3 Code thread #11654 — recurring autonomous runs in one thread
• feat(automation): schedule recurring agent tasks #11815 — recurring agent tasks
• [Feature]: Scheduled Prompts #6748 — scheduled prompts

This overlaps those proposals intentionally but adds the concrete issue-follow-up workflow, change-only reporting, and the distinction between monitoring, drafting, and authorized external replies. Happy for maintainers to consolidate it into the existing scheduling discussion.

[juliusmarminge]

Duplicate of #6748 — that Ideas discussion is the canonical tracker for scheduled prompts / recurring agent tasks on existing threads, including cadence, run history, cross-device execution, and permission/workflow requirements. The linked #11654 and #11815 were already closed as duplicates of #6748. Please add any issue-monitoring or approval-scope details there so scheduling remains consolidated.