gpt-pro-context-handoff

GPT Pro Context Handoff

A coding agent skill for handing off curated project context to ChatGPT Pro through a plaintext HTTPS endpoint, then driving the chatgpt.com session with Browser Use.

Lead with plain readable text over HTTPS, not ZIP upload and not ZIP-primary links. Archives are backups; the reliable handoff is a .txt/.md context file Pro can browse and read directly.

Works with Codex and Claude Code.

What It Does

The skill turns your coding agent into a context broker for GPT Pro reviews, audits, proposals, implementation plans, risk assessments, and strategy analyses. It walks through:

1. Reconstruct The Request - Capture the desired Pro output, project goal, evidence, and constraints
2. Build A Plaintext Context File - One self-contained gpt-pro-context-<timestamp>.txt covering request, orientation, current state, evidence, excerpts, and questions
3. Serve Context Over HTTPS - Random-token path, text/plain; charset=utf-8, verified with curl
4. Prompt GPT Pro - Open chatgpt.com via Browser Use, point Pro at the plaintext URL
5. Wait And Monitor - Watch Pro's visible state and server logs; do not interrupt thinking
6. Capture And Act - Reconcile Pro's response with user constraints; follow up with smaller plaintext files when needed

Install

Manual install

git clone https://github.com/adrianleb/gpt-pro-context-handoff-skill.git

# Codex (global)
cp -r gpt-pro-context-handoff-skill/skills/gpt-pro-context-handoff ~/.codex/skills/

# Claude Code (global)
cp -r gpt-pro-context-handoff-skill/skills/gpt-pro-context-handoff ~/.claude/skills/

# Or copy into a specific project
cp -r gpt-pro-context-handoff-skill/skills/gpt-pro-context-handoff your-project/.claude/skills/

As a Claude Code plugin

claude plugin marketplace add adrianleb/gpt-pro-context-handoff-skill
claude plugin install gpt-pro-context-handoff

Usage

Trigger the skill with natural language:

> package this for GPT Pro
> ask Pro to audit this codebase
> serve project context for a Pro review
> follow up with Pro about <topic>

The skill expects a Browser Use environment for chatgpt.com interaction and an HTTPS host (or local tunnel) able to serve static plaintext.

Key Principles

1. Plaintext over archives - GPT Pro should never need to unzip anything to start
2. Project-neutral by default - The skill itself ships no private hosts, paths, or domain assumptions
3. Verify the fetch, not the 200 - Confirm Pro actually read the context, by visible response and server logs
4. Preserve user framing - Do not silently turn an execution request into a generic strategy memo
5. Do not interrupt thinking - Wait through Pro's thinking/browsing/finalizing states

Project Structure

.claude-plugin/
  plugin.json                             # Plugin manifest
skills/
  gpt-pro-context-handoff/
    SKILL.md                              # Main skill definition
    agents/
      openai.yaml                         # Codex agent surface metadata

Requirements

• Codex or Claude Code (or any coding agent that supports skills)
• A Browser Use-compatible browser automation surface for chatgpt.com
• An HTTPS host or tunnel that can serve static plaintext at a random path

Contributing

Open questions worth discussing:

• What hosting recipes deserve to live in the skill vs. stay user-supplied?
• How should the skill handle multi-file evidence beyond a single .txt?
• What recovery patterns are missing from the current playbook?

PRs and ideas welcome. Open an issue to discuss before large changes.

License

MIT