Skip to content
Back to home

Reliable steering and stop

Lets you send successive steering prompts during a running conversation and still stop the agent afterward. Each sent message is acknowledged independently so one steer does not leave the composer stuck. Stop resolves the currently active Codex turn, including live child work, instead of relying on a stale turn reference.

Branch
GitHubfix/repeated-steering-and-stop
PR
…
Status
Superseded
Reason
Upstream merged Orchestrator V2, which reworked this area, so this branch has nothing left to add. pingdotgg/t3code#2829

Comments

This fork

Extract the sidebar archive actions into a shared hook.

10:42

On it. Pulling the archive handlers out of Sidebar.tsx first, then the header.

Working

Ask anything...

CodexAgent

Composer ready. Agent still running.

Original

Extract the sidebar archive actions into a shared hook.

10:42

On it. Pulling the archive handlers out of Sidebar.tsx first, then the header.

Working

Ask anything...

CodexAgent

Composer ready. Agent still running.

Mock conversation. Steer sends the next follow-up into both running turns; Stop asks both to interrupt. The original's composer waits for an acknowledgement that never matches, and its Stop targets a turn reference that went stale after the first steer.