This project talks to seven commercial AI APIs on a schedule. That means it handles API keys, and API keys are the only thing here worth stealing. This document says exactly how they are handled and how to tell us when something goes wrong.

Locally. Keys live in a .env file that is never committed. .gitignore excludes .env and every .env.* variant except .env.example, which contains variable names only. If you find yourself pasting a key into any file that is not .env, stop.

In CI. Keys are GitHub Actions repository secrets, injected as environment variables into the weekly benchmark job only. The pull-request workflow (ci.yml) references no secrets at all, which is what makes it safe to run on pull requests from forks.

In code. Exactly one module reads provider keys from the environment: src/env.ts. Adapters receive credentials through an injected AdapterContext and are forbidden by an ESLint rule from touching process.env. Nothing under src/providers or src/runner can read a key from ambient state even by accident.

In output. Run files under data/ contain model answers, token counts, and timings. They never contain request headers. Retry logging deliberately omits headers so an Authorization or x-api-key value cannot reach a log. Recorded test fixtures are redacted when captured, and a test scans the fixture directory for key-shaped strings.

In the published site. The site is static. It holds no keys and makes no requests to any provider. The deferred "run your own benchmark" feature keeps a visitor's keys in browser session storage and forwards them per request to a proxy that never persists or logs them; that design is documented in docs/proxy.md before any of it is built.

Email nick@endash.us with hotdogbenchmark security in the subject. Please include what you found, how to reproduce it, and what you think the impact is. You will get an acknowledgement within seven days.

Please do not open a public issue for a vulnerability. Do open a public issue for anything else.

If you believe a key has been committed to this repository or exposed in a workflow log:

1. Revoke the key in the provider's console first. Rotation beats investigation.
2. Email the address above so the secret can be replaced and the history checked.
3. If it is your own key in your own fork, revoke it and rotate — a key in git history is compromised even after the commit is removed.

In scope: this repository's code, workflows, data, and the proxy under proxy/ if present.

Out of scope: vulnerabilities in the AI providers' own APIs (report those to the vendor), and the accuracy of any claim this project makes about whether a hot dog is a sandwich.