handprint.sh · proof of humanity · open protocol

The first art was a human imprint.

When the volume of code is no longer the evidence of human impact, what we can still measure is the provenance of judgment — the decisions, the predictions, and the moments where a person steered the work. handprint.sh is an open protocol for recording that provenance, kept private by default and made auditable on your terms. It exists to define what it means to build, as a human, in the company of machines.

watch it work →see a profilefree · local-first · v0.1.0-draft
human imprint
cueva de las manos · 9,000 bce
or el castillo · 40,800 bp
drop image · jpg/png · 4:5
plate i. pigment blown around a hand pressed to limestone. the earliest authored mark we have — evidence that a particular person had been here.
§ demo ambient capture · local-first

it runs in the background.

handprint watches the seam between you and the tools you already work with — chat, terminal, IDE, team channels — and quietly flags the moments where your judgment shaped the outcome. The session below is a real-time replay from an auth-service-v2 redesign.

Detection runs on a lightweight model on your machine. Your work never leaves your laptop until you choose to seal a record — and even then, every entry stays private until you explicitly mark it public. Every sealed entry is a structured intent record per the handprint protocol v0.1, with cryptographic provenance baked in.

§ five the protocol · open · v0.1

five shapes of steering.

handprint listens for five shapes of human judgment in the work you’re already doing. The structure of each entry — type, intent, risk, horizon — is inferred from the context, and you confirm with a keystroke when something feels worth keeping.

01→ direction“Switching the pipeline to streaming so customer-facing latency drops below 100 ms.”The goal the work moves toward.
02⤴ override“Not the centralized gateway — we tried that in 2024 and the cost wasn’t worth the simplicity. Going with edge JWT.”Choosing your own approach over the AI’s.
03⌀ rejection“We aren’t adding a recommendations engine in v2 — the signal quality isn’t there yet, and neither is the team’s time.”Deciding not to build something.
04▮ constraint“Third-party auth providers cannot sit inside the billing code path — a line we’ve moved twice and won’t move again.”A boundary that has to hold across the system.
05◇ wager“Token sizes will stay under 4 KB across twelve months of growth. We’ll know by May 2027.”A prediction with a date by which you can be held to it. Resolved later as validated, partial, revised, or wrong.
§ profile public · auditable · yours

not a portfolio of outputs.
a track record of thinking.

each handprint is timestamped before the outcome. each resolution is appended, never edited. over months and years a pattern emerges — how this person decides, where their judgment is sharpest, and how honestly they learn when they were wrong.

the centerpiece is the calibration score — computed, never claimed. when this engineer says “i think we should do x,” how much should you trust that?

@cameron · staff infracalibration 0.74
overrideedge JWT on workers over centralized gateway.
11 may · P6M · 0 att
OPEN
due 11/11
rejectiondeclined recs engine in v2. wrong stage.
28 apr · P3M · 4 att
VALIDATED
14d early
wagerp99 under 80ms under 4× load.
14 mar · P1Y · 2 att
PARTIAL
0.6
constraintno third-party auth in billing. ever.
28 feb · ongoing · 3 att
VALIDATED
§ install free for developers · forever

leave your first mark.

Installation is a single command and each tool integration is opt-in. Your handprints live on your machine by default; syncing to handprint.sh is opt-in too, and entries you keep private remain encrypted at rest — only the records you explicitly mark public become searchable.

$curl -fsSL https://handprint.sh/install | sh
brew · npm · standalone binary  ·  see what a profile looks like →