# Chatgpt Pro Review > Use when the user asks for ChatGPT Pro review, external second-pass review, plan hardening, implementation review, PR review, code-review comment resolution, or review of another agent's work. - Skill: `heynag/chatgpt-pro-review` (Agent Skill) - Install (CLI): `npx skillmds@latest add heynag/chatgpt-pro-review` - Raw SKILL.md: https://api.skillmd.com/api/skills/heynag/chatgpt-pro-review/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - License: MIT - Author: heyNag (https://skillmd.com/u/heynag) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/heynag/chatgpt-pro-review --- # ChatGPT Pro Review Use this skill to prepare a focused context packet for ChatGPT Pro, get a second-pass review, and verify the response locally before acting on it. ## Principles - Treat ChatGPT Pro as the reviewer, not the builder. - Treat the current local agent as the reconciler. Do not blindly accept review feedback. - Keep the packet scoped to the decision under review: plan, diff, PR, code-review response, test failure, or eval/reporting method. - Redact secrets, auth tokens, customer data, unrelated private files, and large irrelevant context. - Before submitting private repo context to ChatGPT, confirm the user authorized sending that specific context. If approval is unclear, use the host's normal user-input mechanism to summarize what the packet contains and offer send, edit, or cancel choices. - Ask before implementing feedback that changes product behavior, architecture, security posture, data handling, or scope. - If no approved browser or ChatGPT transport is available, prepare a copy/paste packet and ask the user to send it manually. ## Transport Choose the available path: - If the current agent has an approved browser/control integration and the user authorized sending the packet, open or reuse ChatGPT and submit the packet. - If the user provides a ChatGPT thread URL, use that thread only when the current task fits its context. Otherwise start a clean thread or ask which thread to use. - If transport is unavailable, show the packet in a fenced block for manual paste, then ask the user to paste ChatGPT Pro's response back. ## Workflow 1. Define the review target. - Identify whether this is plan hardening, implementation review, PR/comment resolution, or eval/reporting methodology. - Write the exact decision ChatGPT Pro should review. 2. Gather local context. - Include branch, base branch, PR or issue links, changed-file list, diff summary, key snippets, plan docs, tests run, known failures, open questions, and repo constraints. - Prefer summaries plus essential snippets over dumping whole files. - Include commands already run and their relevant outcomes. 3. Build the review packet. - Use one prompt template from this skill. - Include the output contract. - Keep the packet self-contained enough for ChatGPT Pro to reason without repository access. 4. Send or hand off. - If browser transport is available and authorized, submit the packet to ChatGPT Pro. - If transport is unavailable or not authorized, show the packet for manual paste. 5. Reconcile the response. - Check every actionable claim against the current repo, diff, tests, PR state, and project conventions. - Classify each item as `FIX`, `DEFER`, `DISMISS`, or `QUESTION`. - Implement `FIX` items only when they are in scope and do not need user approval. 6. Report back. - Include ChatGPT Pro's verdict. - Summarize what was fixed, deferred, dismissed, or left as a question. - List local verification commands and results. ## ChatGPT Output Contract Ask ChatGPT Pro to respond in this shape: ```text Verdict: SIGNED OFF | BLOCKED Required changes: - ... Risks: - ... Tests or verification: - ... Reasoning notes: - ... ``` `SIGNED OFF` means ChatGPT Pro found no blocking issue for the stated decision. `BLOCKED` means it found at least one required change or unresolved question. ## Prompt Templates ### Plan Hardening ```text You are ChatGPT Pro reviewing this plan before implementation. Decision to review: [exact decision] Context packet: [branch, issue/PR links, plan, constraints, risks, open questions] Use this output contract: - Verdict: SIGNED OFF or BLOCKED - Required changes - Risks - Tests or verification - Reasoning notes Focus on missing decisions, weak assumptions, sequencing risk, and whether the plan reaches the stated workflow goal. ``` ### Implementation Review ```text You are ChatGPT Pro reviewing this implementation for ship readiness. Decision to review: [exact decision] Context packet: [branch, diff summary, changed files, key snippets, tests run, known failures, repo patterns] Use this output contract: - Verdict: SIGNED OFF or BLOCKED - Required changes - Risks - Tests or verification - Reasoning notes Focus on correctness, edge cases, regressions, missing tests, and whether the diff is shippable for the stated scope. ``` ### PR Or Review Comment Resolution ```text You are ChatGPT Pro reviewing whether PR or code-review comments were resolved correctly. Decision to review: [exact decision] Context packet: [PR URL, branch, review comments, responses, fixes, relevant diffs, tests run, unresolved questions] Use this output contract: - Verdict: SIGNED OFF or BLOCKED - Required changes - Risks - Tests or verification - Reasoning notes Focus on whether any response or fix is incomplete, inconsistent with the repo, or likely to fail review. ``` ### Eval Or Reporting Methodology ```text You are ChatGPT Pro reviewing this evaluation or reporting methodology. Decision to review: [exact decision] Context packet: [metrics, datasets/reports, method, assumptions, comparison criteria, known limitations, relevant code or queries] Use this output contract: - Verdict: SIGNED OFF or BLOCKED - Required changes - Risks - Tests or verification - Reasoning notes Focus on comparability, confounders, missing evidence, and whether the recommendation follows from the reported data. ``` ## Reconciliation Report Report back in this shape: ```text ChatGPT Pro verdict: SIGNED OFF | BLOCKED Reconciled actions: - FIX: ... - DEFER: ... - DISMISS: ... - QUESTION: ... Local verification: - ... Traceability: - Prompt summary: ... - Response summary: ... ```