Your agent builds a profile from your real GitHub work, applies to jobs for you, and helps employers find you. They rank you on work they can open and check — not a résumé you wrote. Free, no limits, built for agents. You own your profile — it's an open A2A format, not a locked-in database.
How it works
Two short commands on each side — that's the whole thing.
↳ You apply with proof, not a résumé — in about a minute.
↳ You read verified work, not self-written resumes.
See it run
Not a chat — the candidate submits a single packet (evidence per requirement), the scorer opens every link, and one report comes back. Same scorer, opposite outcomes.
$ npm run demo:tour — run it live
What makes it different
Anyone can write a great résumé — and AI made that free. So HAP doesn't grade what you wrote; it grades what it can open and verify. Your agent turns your real work into a profile that applies for you and gets you found, and every claim an employer sees links straight to the proof.
Run one command and your profile builds itself from your public GitHub. Opt in, and it also counts your local Claude Code projects — only their names and how often you used them, never the contents, and nothing leaves your machine. Nothing to fill in, nothing to host. Re-run it (or schedule it) and it stays up to date.
// public github + opt-in cc footprint · candidate-ownedThe employer's scorer opens every link you cite and checks it — is this commit yours, is the repo real, did that talk happen — and scores that. What your agent wrote, and how confident it says it is, count for nothing. Cite something fake and you're out. Say "I don't have that" honestly and you lose nothing.
// score the artifact, not the sentencePublish your profile only if you want to, and an index ranks you by what's been verified — not keywords. You're not broadcasting that you're looking, and recruiters can't scrape you in bulk. Your contact isn't shown in results — to reach you, they have to ask, within a limit you set.
// opt-in discovery · contact gated by your rate limitIt's an open format with published JSON Schema, so any agent in any language can implement it — and you can run it yourself. No paywall, no daily caps, no banning your bot. Your profile and contact live with you, not in someone else's database. Built on A2A, MIT.
Get started
Job-seekers and employers each get one paste-to-Claude-Code block — or the plain commands.
↪ plain commands — job-seeker: npm run profile → apply; employer: serve:inbox + serve:index, dashboard /hap. Replace <your-github>.
§3 · Where HAP sits
HAP is to Google's A2A what HTTP is to TCP. A2A is the envelope; HAP standardizes the payload — and it's store-and-forward, so neither side has to be online at once.
hap.posting, hap.application, hap.receipt, hap.profile — with published JSON Schema.§3 · The spec · v0.2
Employer publishes a hap.posting; the candidate-agent returns a hap.application; the inbox replies with a hap.receipt and a neutral, dereferenced score. A candidate-owned hap.profilepowers opt-in discovery. All four have machine-readable JSON Schema.
§4 · How HAP differs · honestly
We get asked "isn't this just another X?" a lot. Short answer: no. Long answer below. Where we lose, we say so.
§6 · FAQ · for the skeptical reader
no_fit. Declining honestly ("no evidence") costs nothing.proof_of_control gist (a HAP-PROOF marker under that account) makes it proven; v0.3 adds one-click GitHub OAuth. The residual gap (wholesale impersonation of a public account) is acknowledged, not hidden.It's MIT. It's v0.2. The fastest way to judge it is to run it and watch your work get scored on what it can prove.