🤖 fix: keep deep research verifiers as leaf agents - #3811
Conversation
Use read-only Explore agents for adversarial claim verification. Allow Explore agents to delegate only to other Explore agents while blocking workflow recursion and full workspace turns. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$16.20`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=16.20 -->
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Workflows to automatically generate PRs for you. |
|
@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: 9bb346c1f2
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".
Derive the Explore task restriction from the full agent inheritance chain so custom read-only agents cannot spawn editing agents or workspace turns. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$22.98`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=22.98 -->
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 32c0eeb36b
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".
Add a workflow-only read-only verifier agent with task and workflow orchestration disabled. Use it for deep-research claim votes instead of relying on prompt instructions. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$27.06`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=27.06 -->
|
@codex review |
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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 participant
I noticed the
deep-researchclaim verifiers could treat their prompt as another research request and start a nesteddeep-researchworkflow. This made a single verifier fan out into another full workflow, which was a lot more work than intended.I initially moved the verifier to
explore, but that agent intentionally keepstaskso it can delegate read-only investigation. The fix is just to add a workflow-onlyresearch_verifieragent which inherits the research tools fromexplore, but cannot usetaskor start workflows. Normal Explore-to-Explore delegation is unchanged, and custom agents which inherit fromexplorekeep the same read-only task boundary.Generated with
mux• Model:openai:gpt-5.6-sol• Thinking:xhigh• Cost:$27.06