Asking to create documentation or specs in Plan Mode can trigger actual implementation instead after confirmation

Resolved 💬 1 comment Opened Jun 21, 2026 by procommerz Closed Jul 9, 2026

What version of the Codex App are you using (From “About Codex” dialog)?

26.616.51431

What subscription do you have?

Pro

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

When using Plan Mode to plan updates to the documentation or tasks, when the plan is confirmed, Codex might try to actually work on the implementation, rather than stopping after making updates to documentation and/or tasks.

I believe, this happens because confirming a plan automatically inserts a user message with text "Start implementation". This message produces a logical collision with the original request to work on the documentation.

Even explicitly specifying "no code changes, only documentation updates" in the original prompt doesn't help sometimes.

And if I use the text input "No..." option entering something like "Yes, create the documentation", the agent will re-enter planning mode, instead of proceeding to the expected action of working with the doc files.

What steps can reproduce the bug?

Enable Plan Mode and ask Codex to "prepare a spec documentation for a simplistic moon lander game and scope it into a five-step implementation plan". Confirm the plan - observe that the agent might actually implement the code, not just the docs.

What is the expected behavior?

After the plan is confirmed, the agent should work only on the documentation files, not the source code.

Alternative: when I use the existing custom negative response to the plan, saying "Yes, implement the documentation", the system should follow my instruction and not re-enable planning mode, like it currently does.

Additional information

Proposed solution:

Basic: add just one more secondary "Yes" option with custom user input (automatically exits plan mode, like the existing "Yes" option)
Elaborate: generate 2-3 alternative "Yes, ..." options based on the wording of the original user's prompt that triggered planning, automatically exiting planning mode and ensuring it's not re-enabled on the next turn.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗