Conversation
|
This change is part of the following stack: Change managed by git-spice. |
![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: 43c27ff146
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".
| "models": [ | ||
| { |
There was a problem hiding this comment.
This catalog drops gpt-5, gpt-5-codex, and gpt-5.1*. spawn_agent validates requested models against list_models() and now rejects those names as unknown instead of accepting existing user configs/automations. That is a breaking behavior change for persisted workflows that still pass legacy slugs.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d851152281
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".
| "models": [ | ||
| { |
There was a problem hiding this comment.
This catalog now omits legacy slugs (for example gpt-5-codex/gpt-5.1-codex). When users keep those configured, get_model_info() falls back to model_info_from_slug, which sets generic metadata (shell_type: default, apply_patch_tool_type: null). That silently changes tool behavior/output formatting for existing installs instead of preserving prior model capabilities.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 83f145f3f5
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".
| @@ -1,9 +1,11 @@ | |||
| { | |||
| "models": [ | |||
There was a problem hiding this comment.
gpt-5.2-codex was a list-visible bundled preset before this change, but it is now absent from the catalog. spawn_agent does an exact match against list_models() and returns Unknown model on misses (find_spawn_agent_model_name in core/src/tools/handlers/multi_agents_common.rs). Persisted configs/automation still using gpt-5.2-codex will now fail instead of continuing to work.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c9bf7c3e90
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".
| @@ -1,9 +1,11 @@ | |||
| { | |||
| "models": [ | |||
There was a problem hiding this comment.
Removing gpt-oss-20b and gpt-oss-120b from the bundled catalog breaks existing spawn_agent workflows that still request those model names. find_spawn_agent_model_name does an exact match against list_models() and returns Unknown model on misses, so persisted configs/automations now fail instead of continuing to run.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0f27d9258b
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".
| @@ -1,9 +1,11 @@ | |||
| { | |||
| "models": [ | |||
There was a problem hiding this comment.
The new catalog removes gpt-5.1-codex-mini, but TUI still hardcodes it as NUDGE_MODEL_SLUG (tui/src/chatwidget.rs:469). lower_cost_preset() only shows the rate-limit switch prompt when that exact slug is present (chatwidget.rs:7573), so users hitting high usage no longer get the lower-cost model nudge.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ce95cf5f56
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".
| "context_window": 272000, | ||
| "max_context_window": 272000, | ||
| "auto_compact_token_limit": null, |
There was a problem hiding this comment.
Every bundled model entry now omits max_context_window. In with_config_overrides, model_context_window is clamped only when this field is present, so user overrides become unbounded. A large configured context window can now bypass the previous 272k limits, delaying compaction and producing oversized requests for these models. Add max_context_window back per model (usually matching the real cap).
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3ead933161
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".
| @@ -1,9 +1,11 @@ | |||
| { | |||
| "models": [ | |||
There was a problem hiding this comment.
Removing gpt-oss-20b/gpt-oss-120b from the bundled catalog forces configured users onto model_info_from_slug fallback (models-manager/src/model_info.rs), which hardcodes max_context_window to 272k and apply_patch_tool_type to None. Those slugs previously had 128k windows and freeform patch support, so long sessions can exceed real limits and patch tooling silently disappears.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1f74ed0c34
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".
| @@ -1,9 +1,11 @@ | |||
| { | |||
| "models": [ | |||
There was a problem hiding this comment.
The catalog now omits legacy gpt-5*/gpt-5.1-codex* entries, so users pinned to those models are no longer present in available_models. In handle_model_migration_prompt_if_needed (tui/src/app.rs), migration metadata is read from the current preset; if absent, the function exits early and never shows the upgrade prompt.
Useful? React with 👍 / 👎.
2 participants
Uh oh!
There was an error while loading. Please reload this page.