ProofLoop 5.6 + Luis Rodriguez Rivera Portfolio

ProofLoop is an evidence-first acceptance workflow for agent-written software. It combines a repo-scoped Codex skill powered through a ChatGPT subscription with a zero-setup browser demonstration of the same domain, security, product, test, and human-decision gates.

• Live project: https://luisrodriguezdev.netlify.app/agent-review-lab/
• Source: https://github.com/luisangelrod/Portfolio
• OpenAI Build Week track: Developer Tools

Why ProofLoop exists

Coding agents can produce a plausible implementation before anyone has proved that its business rules, tenant boundaries, failure states, or regression coverage are correct. ProofLoop adds the control loop after generation:

1. Bound the goal and supplied evidence.
2. Apply independent domain, security, product, and test review gates.
3. Label claims as observed, executed, proposed, or unknown.
4. Convert findings into a bounded correction and regression plan.
5. Reserve the final accept/reject decision for a human.

Use it with a ChatGPT subscription

No OpenAI API key is required.

1. Install the current Codex CLI.
2. Run codex login and choose Sign in with ChatGPT.
3. Clone this repository and start Codex from the repository root.
4. Invoke the checked-in skill:

$proofloop-review review this diff and decide whether it is ready for human acceptance

Codex discovers the skill at .agents/skills/proofloop-review/SKILL.md. It inspects relevant source and tests, runs proportionate checks, cites evidence, and separates executed verification from proposed work.

Judge test path (under one minute)

1. Open the live ProofLoop lab.
2. Choose any of the four failure boundaries.
3. Select Run judge demo.
4. Inspect the four verdicts, correction plan, and regression evidence.
5. Make the required human decision and download the JSON evidence report.

To test the real GPT‑5.6 workflow, select Copy Codex review prompt and paste it into Codex from this repository. The browser demo is explicitly labeled as curated evidence and never pretends to be a live model response.

Technical design

• Repo-scoped Codex skill for the subscription-native GPT‑5.6 workflow
• Semantic HTML, modern CSS, and vanilla JavaScript for the hosted judge demo
• Four curated regression scenarios with deterministic evidence output
• Keyboard-accessible tabs, reduced-motion support, human decision lock, and JSON export
• Optional server-side Responses API adapter retained for teams that separately choose API billing; it is not enabled or required by the public submission

Validate locally

Requirements: Node.js 20+. No credential is needed for the public demo.

npm install
npm test
npm run build

For local hosting, run npm run dev. The static site is also deployable directly with netlify.toml.

Submission materials

• Submission copy
• Under-three-minute demo script
• Architecture and safety notes
• License and asset exceptions

Other portfolio routes

• / — professional portfolio
• /agent-review-lab/ — ProofLoop 5.6
• /eval-flight/ — interactive evaluation mission
• /desktop-1995/ — Windows XP memory desktop alternate

Author

Luis Rodriguez Rivera — GitHub · LinkedIn · Email