chatgpt-browser-oracle

Automates approved ChatGPT browser sessions to submit prompts and download generated artifacts.

Updated Apr 14, 2026
One-click install
npx skills add https://github.com/amitkarpe/agent-skills --skill chatgpt-browser-oracle-amitkarpe
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: chatgpt-browser-oracle
Source: https://github.com/amitkarpe/agent-skills/tree/main/skills/chatgpt-browser-oracle
Command: npx skills add https://github.com/amitkarpe/agent-skills --skill chatgpt-browser-oracle-amitkarpe

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires chrome-remote-interface, and includes scripts (resource) and references (resource) components.

What problem does it solve? This Skill handles the narrow case of running a legacy, explicitly authorized ChatGPT browser workflow: submitting a prompt to a ChatGPT project through a real Chrome session, confirming the requested model mode, downloading the generated artifact, and validating the file before reporting success. ## Core Features & Use Cases - Browser session control: Focuses the Oracle Chrome control tab and connects to the DevTools endpoint at 127.0.0.1:9222 via chrome-remote-interface. - Model verification: Inspects or selects Instant, Thinking, or Pro modes and saves JSON proof of the visible composer model state. - Artifact download and validation: Downloads the newest matching generated file, verifies SHA-256 hashes across copies, and validates ZIP, DOCX, PPTX, XLSX, PDF, HTML, JS, image, and GIF outputs. - Use Case: An owner approves a one-off task to generate an XLSX report through a specific ChatGPT project; the Skill submits the prompt file, confirms Pro mode, downloads the workbook, and validates xl/workbook.xml before returning the artifact path. ## Quick Start Ask the agent to run the approved ChatGPT browser artifact workflow with your prompt file and output directory, then verify the downloaded file with the validation script.

Frequently Asked Questions about chatgpt-browser-oracle

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate ChatGPT in a browser using Chrome DevTools Protocol?▼

Connect to a running Chrome instance with chrome-remote-interface on the DevTools port, then use Runtime.evaluate to find the prompt composer, insert text, and dispatch Enter. The scripts here target 127.0.0.1:9222 and a chatgpt.com page.

How do I verify which ChatGPT model mode is selected?▼

Run chatgpt_ensure_model.js in inspect mode to read the composer button text and message model slugs from the DOM. With --select true and a mode of instant, thinking, or pro, it opens the model menu and clicks the matching item, saving JSON proof.

Can this Skill be used for GitHub repository reviews or PRs?▼

No. The SKILL.md scope boundary explicitly forbids using this browser path for GitHub reviews, Issues, PRs, or ChatGPT-Codex collaboration. Those tasks must use the owning repository's collaboration protocol instead.

How do I validate a downloaded DOCX or XLSX file from ChatGPT?▼

Run validate_generated_file.sh with --expect docx or xlsx. It checks the ZIP integrity with unzip -t and requires the OOXML root entry, word/document.xml for DOCX or xl/workbook.xml for XLSX, plus a SHA-256 hash record.

Why does the ChatGPT download script fail to find the file?▼

The script polls the Downloads directory for a new non-temporary file matching the expected extension and name within the timeout. Failures occur when the download button selector changed, a save dialog blocked the download, or the file landed outside the configured downloads directory.

What are the limitations of browser-based ChatGPT automation?▼

It requires explicit per-task owner approval, a running Chrome user service, and a live DevTools endpoint. Standalone JS, PNG/JPEG, and video downloads are not yet validated in this harness, and stale script defaults must be checked against the approved target.