I Split the Work Between Codex and ChatGPT Pro: One Creates, One Verifies

The problem: when two AIs work together, who's in charge?

I use Codex for coding, but sometimes I also need ChatGPT Pro for image generation, long document drafts, or strategic research. The problem is there's no natural boundary between the two AIs — it's easy to end up with Codex accepting Pro's output without verification, or sensitive information leaking across the wire.

The 365-chatgpt-pro-delivery skill solves this boundary problem.

Core design: three lanes, smallest brief, single dispatch

The skill splits collaboration into three lanes, chosen by evidence need, not apparent complexity:

• FAST_DELIVERY — images, copy, document drafts, supplied-evidence synthesis; no new external facts needed.
• RESEARCH_DELIVERY — current or specialist external facts can materially change the decision; visible sources and dates required.
• ENGINEERING_DELIVERY — Codex expects source changes, file-specific review, test ideas, or runtime guidance; supply only the narrow context needed for correctness.

Every dispatch includes a smallest complete brief: Goal, Lane, Evidence supplied, Evidence not supplied, Required output, Hard constraints, Acceptance criteria. Not one word more.

Evidence labels: Pro output is only a candidate

Pro must label its claims:

• DIRECT — supported by supplied or visibly inspected evidence
• INFERRED — reasoned but not directly confirmed
• ILLUSTRATIVE — example, concept, mockup, or pseudocode
• UNVERIFIED — not confirmable from available evidence

Release-blocking claims require DIRECT. Pro must not claim it ran commands, opened files, or tested runtime behavior it could not access.

Codex accepts independently, never blindly trusts

Codex checks against the user's goal and real local evidence:

• ACCEPTED_UNCHANGED / ACCEPTED_WITH_LOCAL_EDIT
• RETURN_TO_PRO (with observed evidence, violated constraint, required correction)
• REJECTED / UNVERIFIED

Small fixes happen locally; new reasoning gets one delta correction, never a resend of accepted context.

Install

bash git clone https://github.com/laojin1900/365Skill.git ./install.sh 365-chatgpt-pro-delivery

Repo: github.com/laojin1900/365Skill/tree/main/skills/365-chatgpt-pro-delivery