Feature request: Orchestrated ChatGPT ↔ Codex workflow with independent review loop

Title: Feature request: Orchestrated ChatGPT ↔ Codex workflow with independent review and shared repository context I would like to suggest a supervised multi-agent workflow that connects ChatGPT and Codex more directly. The current manual workaround is something like this: 1. I describe a coding task to ChatGPT. 2. ChatGPT turns it into a precise Codex instruction. 3. I paste that instruction into Codex. 4. Codex produces a patch or diff. 5. I bring the diff back to ChatGPT for review. 6. ChatGPT finds issues, risks, regressions, UX concerns, or missing tests. 7. I paste the follow-up instructions back into Codex. 8. This loop continues until the patch looks safe enough to apply. This workflow works surprisingly well, but it is slow and repetitive because the user has to manually carry context, instructions, patches, and review comments between ChatGPT and Codex. Another important part of this request is repository access. Ideally, ChatGPT, acting as the planner/reviewer, should be able to access the same user-approved GitHub repository or local workspace that Codex uses. Today, a common workaround is to upload a project archive to ChatGPT, ask it to inspect the project, ask it to write Codex instructions, then manually move patches and review comments back and forth. That is not ideal. The planner/reviewer needs real project context: file structure, existing code, previous changes, diffs, tests, documentation, and project-specific instructions. Codex also needs to run in that same repository or workspace so that implementation and review are based on the same source of truth. A better workflow would be: 1. The user grants ChatGPT and Codex access to a specific repository or workspace. 2. ChatGPT can inspect the repository in read-only mode by default for planning and review. 3. ChatGPT can start a Codex task inside that repository. 4. Codex creates a branch, worktree, patch, or pull request. 5. ChatGPT reviews the resulting diff using the same repository context. 6. Codex addresses the review comments. 7. The loop repeats until the reviewer approves the patch or asks the human for a decision. 8. The user approves or rejects the final patch. The user should not have to repeatedly upload project archives or manually copy diffs between ChatGPT and Codex when both tools are part of the same ecosystem. The key point is role separation. I do not want the same coding agent to be the only reviewer of its own work. Self-review is useful, but it is not the same as an independent review step. Ideally, the implementing agent and the reviewing agent should have clearly separated roles, instructions, and possibly context. This would be especially useful for users who are not professional programmers but can still validate product behavior, UX, workflows, and edge cases. Such users can describe what they need and test the result, but they may need help turning ideas into safe code changes and reviewing diffs before applying them. This request is related to existing discussions about Codex PR review, ChatGPT Projects using local folders, Ask/Patch mode, closed-loop validation, and multi-agent workflows. The missing piece is a first-class bridge between ChatGPT-style planning/review and Codex-style implementation. Important safety requirements: * Destructive actions should require explicit approval. * Production changes should require explicit approval. * Sending messages, deleting data, submitting forms, or applying final patches should require explicit approval. * The system should show the diff and review result before anything is applied. * The user should be able to stop the loop at any point. * ChatGPT/reviewer access should be read-only by default unless the user explicitly allows a broader action. In short: I would like a safe, supervised coding loop where ChatGPT can act as planner/reviewer with access to the user-approved repository context, Codex can act as implementer in that same repository, and the human remains the final approver.