Agent should execute tasks end-to-end without stalling on optional prompts

[likeyiyy]

Problem

In real work, users assign a task and expect the agent to continue autonomously to completion.

However, the current behavior often stalls on optional phrases like:

• "if you'd like"
• "let me know"

This creates a stop-and-wait workflow and breaks the expected autonomous execution model.

Expected behavior

1. Action-first execution by default (analysis -> code changes -> verification -> delivery).
2. Ask users only when truly blocked by missing required input or risky irreversible action.
3. Avoid optional invitation tails that pause execution.
4. Keep progressing until the task is objectively complete.

Why this matters

For modern agent workflows, users should be able to assign a task, leave for an hour, and come back to a completed result instead of a stalled suggestion.

Context

This issue is about default interaction behavior and completion semantics, not model capability limits.

[github-actions]

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• Creating a skill + automation does not trigger immediate execution #12395
• Remove collaboration_modes runtime feature gating #12018
• Agent loop fails to start after task recognition (agent roop stops and waits for input) #11527

Powered by Codex Action

[etraut-openai]

Thanks for the bug report. We'll share the feedback with the team responsible for training our codex models. We're tracking this feedback and similar reports so we can improve future models. And thanks for using the /feedback command. That additional information is useful.