Feature request: one-click ChatGPT → Codex handoff with bidirectional task/results flow #40142

Problem

ChatGPT and Codex are currently powerful together, but the handoff still makes the user act as a manual API between two OpenAI products.

A concrete example is environment setup for an existing GitHub project:

1. The user discusses the project and constraints in ChatGPT.
2. ChatGPT determines exactly what Codex should verify or configure.
3. The user manually copies the prompt into Codex.
4. Codex runs the task and returns a diagnostic.
5. The user manually copies the diagnostic back into ChatGPT.
6. ChatGPT analyzes it and produces the next prompt.
7. The user copies that back into Codex again.

This loop is especially painful for technically capable users who are not professional developers. The user should not need to understand which product should receive which instruction or repeatedly shuttle text between them.

Desired experience

ChatGPT should be able, with explicit user approval, to create or continue a Codex task directly and receive its structured result back.

Ideal flow:

User discusses need in ChatGPT
        ↓
ChatGPT prepares the Codex task
        ↓
User approves “Send to Codex”
        ↓
Codex executes in the selected repository/environment
        ↓
Result, blockers, diff, tests and status return to ChatGPT
        ↓
ChatGPT evaluates the result and can propose the next Codex action

The same mechanism should work for:

• initial Codex environment configuration
• repository audits
• implementation tasks
• test failures
• code review
• follow-up corrections
• deployment preparation

Important product boundary

This should not mean silently giving ChatGPT unrestricted execution rights.

The user should explicitly control:

• which repository/environment Codex may use
• whether the task is read-only or may modify files
• whether network access is allowed
• whether secrets are available
• whether Codex may create a branch/PR
• whether deployment is allowed

The handoff should preserve these permissions rather than forcing the user to restate them in every prompt.

Why this matters

ChatGPT often has the broader product discussion, requirements, screenshots, prior decisions and user intent. Codex has the execution environment and repository tools. These are complementary roles.

Today, the missing bridge means the human becomes glue between them. For non-developers using AI to build real software, this is a major usability barrier and a source of mistakes because prompts, diagnostics or constraints can be copied incompletely.

A first-class handoff would let ChatGPT act as the planning/supervision layer and Codex as the implementation/execution layer, while keeping the user in control of permissions.

Suggested MVP

1. A Send to Codex action from a ChatGPT conversation.
2. Repository/environment selector.
3. Permission summary before launch.
4. Ability to attach the relevant ChatGPT context or a generated task brief.
5. Codex task status visible from ChatGPT.
6. Structured return of final status, changed files, tests, blockers and PR link.
7. Continue in ChatGPT and Send follow-up to Codex without copy/paste.
8. Preserve an explicit audit trail of what ChatGPT sent and what Codex executed.

Related

This overlaps strongly with #33249, especially the need for ChatGPT to supervise active Codex tasks. This request emphasizes the end-to-end user workflow: direct task creation, environment/configuration handoff, structured results returning to ChatGPT, and eliminating manual copy/paste in both directions.