Somtimes the agent does something the wrong way.

Somtimes the agent does something the wrong way. e.g. open a PR with the `gh` CLI instead of through whatever internal mechanism.
It doesn't seem like there is a good way to recover from this other than to have it spin up a new session from the created PR.

---
| Field | Value |
| --- | --- |
| App version | 1.0.23 |
| OS | NixOS 26.5.0 |
| Theme | GitHub |
| Path | /chat |
| Tenure | Week 7 |

Start at the /chat entry point and trace how the agent creates a pull request when it uses the gh CLI instead of the app's internal mechanism. Compare the existing PR flow with the reported recovery path from a created PR, then define a reproducible failure case and the expected recovery behavior.

Written by the indexing model from the issue text.