Collapse mechanical continuation prompts without obscuring human requests

[sunlover7]

Automated continuation prompts can dominate a thread when scheduled work or quota recovery repeatedly posts a full brief. They are still full user-role bubbles even when attributed to automation. The Working section helps with the sidebar, but does not reduce transcript clutter. Reducing unnecessary prompt production is useful separately; this proposal concerns presentation of messages that actually exist.

I would like maintainer direction before opening a feature PR, per CONTRIBUTING.md.

Proposed scope:

• Fold adjacent, positively identified scheduled prompts and quota-reset continuations into a counted Background activity disclosure.
• Retain original bodies, timestamps, sender/task links and message/run identities on expansion. Keep search and minimap navigation usable.
• Never classify from wording such as “Cont” or “Pending.” Actual human messages remain prominent. Generic agent requests and unknown-origin messages stay expanded.
• Never group across an assistant outcome, human message, failure, permission request or tool event. This should not hide actionable decisions or consent.
• Treat folding as presentation only. It does not remove model input or reduce the cost of a turn already dispatched.

A working prototype is available at sunlover7/t3code, commit 1fc912c3. It is based on the installed 0.1.0 source, 81493efb71e8e8244f95d3e9f71d8369654ab151, rather than proposed as a merge-ready feature against current main. It includes shared provenance handling, web/desktop adjacent-event grouping, and individual disclosures on mobile. The default/opt-in choice is deliberately a maintainer decision before a feature PR.

The prototype passed focused chat/shared/recovery tests, web/server typechecks and a web production build. An isolated React disclosure example was checked in light/narrow and dark/wide layouts with expand/collapse detail retention. Native mobile execution and installed-client rollout were not verified; no live app was replaced. A smaller backend quota-provenance defect can be addressed independently through the existing bug-fix route.

Would this direction fit T3, and should it be an optional preference or the default for these known mechanical triggers? I will keep the presentation feature out of a PR until the direction and scope are approved.

Prototype written with GPT-6.1 Sol through the Codex harness.