Keep your coding agent on task when you add requests
A developer wrote a skill that makes Codex and Claude Code finish the current job before new requests; one live Codex test passed all 10 checks, they report.
Done with OpenAI Codexgpt-6-sol
What they did
A developer wrote a set of standing instructions, called a skill, for Codex and Claude Code. When you send a new message mid-task, the agent decides whether it is a correction, a question, or a new job. New jobs wait in a queue until the current one is finished and checked.
How it went
The author reports a live Codex test where a second request arrived mid-task. The first task was finished and checked before the second began, passing all 10 checks. A stop test passed 11 of 12, but cancelling took about 35 seconds.
Worth knowing
It is only instructions, so it cannot force the agent to comply. Claude Code was not part of the live tests.
Try it yourself with OpenAI Codex
For the rest of this session, follow this rule. Keep one active task at a time. If I send a new, unrelated request while you work, add it to a queue and tell me it is queued. Then finish and check the current task before starting it. If my message corrects the current task, apply it right away. If I ask for status, answer briefly and continue. If I say pause, cancel or stop, do that at once and don't start queued work. Show me the queue whenever it changes. Start with: [your first task].Swap the highlighted [parts] for your own details.
Source: X · @emreba4g · Oct 17, 2026
Written with AI assistance from the original post and checked against our editorial guidelines. Results are as reported by the source, not independently verified.
