feat(personality) introduce model_personality config#9459
feat(personality) introduce model_personality config#9459dylan-hurd-oai merged 6 commits intomainfrom
Conversation
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 64e0e9d393
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| let template = include_str!( | ||
| "../../templates/model_instructions/gpt-5.2-codex_instructions_template.md" | ||
| ); |
There was a problem hiding this comment.
When model_personality is configured, this always loads the GPT‑5.2 Codex template and (a few lines below) replaces model.base_instructions with it. That means any non‑GPT‑5.2 model (e.g., o3, gpt-4.1, or codex-mini) loses its model‑specific prompt content such as BASE_INSTRUCTIONS_WITH_APPLY_PATCH, so users who enable personality while using those models will get mismatched GPT‑5.2 wording and may lose tool guidance. Consider injecting the personality into the existing model instructions (or selecting a template per model) instead of unconditionally switching to the GPT‑5.2 template.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
I'm going to leave this as a follow-up for now, to resolve once #9427 lands. It should make it simpler to manage this
There was a problem hiding this comment.
This is fine for now - going to defer until we have a proper templating system in place.
dylan-hurd-oai
force-pushed
the
dh--personality
branch
from
6811830 to
fa7d0ad
Compare
2 participants
Summary
Introduces the concept of a config model_personality. I would consider this an MVP for testing out the feature. There are a number of follow-ups to this PR:
Testing