Repository navigation
Prefer balanced current models and native Claude roles - #5
Conversation
![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: 5746fb2ddc
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".
| options=merged, model=spec.model, | ||
| model_policy=self.settings.model_policy, |
There was a problem hiding this comment.
When an embedding upgrades with a native callback written for the previous five-key contract (role, modality, system, contents, and options), these unconditional new keyword arguments make every native invocation fail with TypeError before the callback runs—even when no model or policy override was requested. Preserve the existing public runtime callback contract, such as by passing the new metadata only when supported or adding an explicit compatibility adapter.
AGENTS.md reference: AGENTS.md:L7-L7
Useful? React with 👍 / 👎.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
PaperVizAgent could inherit an expensive host model or the Codex account default for routine roles. Prepare 0.5.0 with one balanced/quality policy across native skills, Python callbacks and every generated operation surface. Claude uses moving Sonnet/Opus aliases; Codex reads the current account catalog, selects from provider tier hints and records its reason, effort and fallback. Explicit model and effort choices remain authoritative, including available hidden/specialized models.
Native text and vision remain in the calling host. Only a missing modality falls back to Codex. Document the supported Claude skill plus Git-source MCP installation while PyPI publication is pending, and use native completion events without scheduled polling.
Validation:
make check: drift check and 52 tests passed (52 subtests).make validate: all upstream host validation stages passed with Node 24, pinned Hermes and OpenClaw 2026.9.2.make package: all 13 packaging steps passed (credential-gated signing skipped when unconfigured).sonnettoclaude-sonnet-5; a fresh agent inspected an existing PNG, and two independent planners completed via notifications without scheduler calls.Image-generation limitation: Codex OAuth exposes a managed image tool with a coordinating LLM request; it neither accepts nor reports the underlying image-model ID. The exact
gpt-image-2requirement is awaiting the user's decision. No new image inference was run or claimed in this validation. Original upstream role prompts and style guides are unchanged.Embedded callback migration: 0.5.0 adds
modelandmodel_policykeyword arguments to native callbacks. Explicit old signatures must be updated; the callback test uses the full explicit contract. Generated adapters and skills already carry the new policy.