Run a configurable prompt or skill when settling a thread

[X2Borgo]

Problem

The Settle thread action only flips thread state. Many of us end a thread with a wrap-up routine (link/unwatch PRs, clean worktrees and temp files, update the tracker, write a hand-off) that we currently trigger by typing a slash command such as /my-settle-skill, which then settles the thread itself. Clicking Settle in the UI skips that routine entirely, so the button and the "real" way to finish a thread diverge.

What exists today, and why it is not enough

• runOnSettle project action (feat(server): run a project action when a worktree thread settles #16290, shipped). A project action with "runOnSettle": true runs a shell script in the thread's worktree when the thread settles. It runs a terminal command, not a prompt or skill, so it cannot invoke a skill inside the thread's own agent session. The only workaround is a script that starts a separate headless agent (for example claude -p "/my-settle-skill"), which has none of the thread's context.
• Per project, not global. It is configured in each project's t3.json; there is no global setting, so "run this on every project" means editing every project.
• Worktree threads only. Threads in a project's main checkout are skipped.
• Related proposals. [Feature request] Run an agent automation/prompt on auto settled thread #16820 asks for an agent prompt to run when a thread is auto-settled after a merge (the same idea, scoped to auto-settle and a Jira example). Add lifecycle hooks for repo threads #13315 (closed) led to runOnSettle. feat(scheduled-tasks): optionally settle threads on completion #15167 is the reverse direction: settle a thread when a scheduled task finishes.

This proposal is the prompt/skill counterpart of runOnSettle, run in the thread itself.

Proposal

An optional settle prompt: a prompt or skill invocation (e.g. /t3-settle) that T3 sends to the agent as a final turn when the user clicks Settle, then settles the thread once that turn completes.

• Configurable globally, with an optional per-project override (the per-project runOnSettle setting already shows where a project-level hook lives).
• Off by default; Settle stays instant when unset.
• The skill should not need to call settle itself: T3 settles the thread after the turn, so a wrap-up skill is not circular.

Open questions

• Running thread: queue the wrap-up turn, or refuse/ask?
• Settle timing: settle immediately and run the wrap-up in the background, or only after the turn finishes (and what if it fails)?
• Auto-settle (on merge / after N days): should it also run the prompt? Unattended turns cost money and may surprise people, so probably opt-in separately or never. [Feature request] Run an agent automation/prompt on auto settled thread #16820 wants exactly this for merges.
• Providers without skill/slash-command support: send the text as a plain prompt, or hide the option?
• Relation to runOnSettle: which runs first when both are set, the prompt turn or the script?
• Reverse state: un-settling is unaffected, but should the UI show that a wrap-up turn ran?