Gpt-5.6-sol doens't do what you ask, it decides to do its own thing

What version of Codex CLI is running?

0.149.0

What subscription do you have?

Pro

Which model were you using?

gpt-5.6-sol

What platform is your computer?

Darwin 25.5.0 arm64 arm

What terminal emulator and version are you using (if applicable)?

zsh

Codex doctor report

What issue are you seeing?

The agent will avoid doing the task it has been given to do, and will create different work that consumes tokens.
I gave the agent 6 specific tasks that he was to work on all night. what i ended up getting was not a single task complete, and a new feature i didnt ask for. furthermore the agent stopped working due to a sandbox bug driving the chrome session that it was driving earlier in the session. this caused a fake blocker, and no work done, and a bunch of code that i didnt ask for.
The behavior with the browser or gh command suddenly being not usable by the agent is a huge problem causing work stoppages

What steps can reproduce the bug?

Uploaded thread: 01a02098-2fde-7820-9c2b-990c67b99f49

What is the expected behavior?

Three thrings. Manage subagents to get the list of tasks accomplished - manage the users list of tasks, and not make up new features or over engineer

Additional information

this behavior has made sol so bad to use, that i cannot trust it anymore.
constantly disobeying the user, ignoring AGENTS.md guardrails that instruct you to not get into this situation we are in.

Start by reviewing the uploaded thread 01a02098-2fde-7820-9c2b-990c67b99f49 and reproducing the report with Codex CLI 0.149.0. The issue names no source file, test, or entry point; completion would require a narrowed, reproducible failure and explicit acceptance criteria for task adherence and sandbox or browser reliability.