Bragproof

How it works, and why

An agent that needs a tool finds a package name and has to decide: will this start, what will it want from me, and what will it do on my machine? Today the only way to know is to run a stranger's code and see.

The idea

Bragproof runs it first, somewhere that does not matter, and publishes what happened. Each server gets a run report: not a review, not a score, and nothing the maker wrote about themselves. Only what a neutral machine saw.

If you are choosing a tool

See whether it starts with no setup, which settings it asks for, what it contacts, and what its tool descriptions will say to your agent, before any of its code runs near you.

If you made one

Type its name. It is listed where agents search, with a report of whether it starts and what it needs, without a launch post or a following. Get listed

What a report records

You do not have to trust this site

A report is made by a public job on a machine GitHub hosts, running this project's open workflow. When it finishes, GitHub signs a token that names the workflow, the commit it ran from, and the hash of the report's exact text. The site stores the text and the token and serves both.

  1. Fetch https://bragproof.smaverk.com/v1/reports/<package>?raw=1. You get the report text and the token.
  2. Hash the text with SHA-256. The token's aud must be bragproof-report: followed by that hash.
  3. Check the token's signature against GitHub's public keys, and that it names itsHaddad/bragproof/.github/workflows/report.yml@refs/heads/main on a GitHub-hosted machine.
  4. Read that workflow at the commit the token names, and the run's public log.

Or let the open-source tool do all four: bragproof check <package> from the repository. What you end up trusting is GitHub and code you can read. GitHub retires its signing keys after a while; an old token then no longer checks, and the report should be run again.

What a report does not tell you

"Needs setup" is not a fault. Many good servers need a key. It is there so you know before you try.

Proven calls

A second, stronger kind of evidence, with three levels

Separately from reports, a maker can prove that specific calls return specific results. Those proven calls carry a receipt at one of three levels.

LevelWhat vouches for the runWhat you are trusting
AttestedA confidential machine (for example Intel TDX): the hardware signs what ran.The chip maker, the cloud host, and that nobody had physical access.
WitnessedGitHub: it signs that this project's check workflow ran on its machine.GitHub. No hardware evidence underneath.
Self-reportedNothing.Whoever ran it. Not shown unless you ask.

Proven N times by N accounts counts passing runs and the different GitHub accounts or cloud projects that asked for them, the maker's own included. Accounts are cheap to make: read two or more as "someone besides the maker checked", not as a vote.

Get your server listed