salluru.dev
operational

All systems active

Labs / Computer-Use
Launching

guiproof

Is this UI automation still true — today, for this version?

Continuously-verified, version-pinned GUI behavioral facts for agents — skip re-discovery every session.

At a Glance
version
pinned
freshness
score
re-verify
flywheel
neutral
schema
The Problem

A published agent skill (“Export is under the chevron”) is a snapshot that silently rots as the app changes — the button moves, a modal appears, a label goes inert. The failure is quiet: the agent exports the wrong report and nobody knows. There's no freshness signal and no version pinning, so serious deployments re-discover every app from scratch.

Key Insight

The value isn't the knowledge — it's the freshness signal on it. UI changes weekly, so a trustworthy, version-pinned freshness oracle is enormously valuable and hard to clone. Crowdsource cheap, noisy agent-run observations; own the re-verification engine that converts them into version-pinned, trust-weighted truth.

How It Works

Decaying confidence

Each behavioral fact carries (app, version, fact, last-verified, confirmations, decay) — a freshness score, not a static tuple.

Re-verification engine

Batched sandbox runs re-confirm facts across tenants; cost-per-fresh-fact drops with scale.

Version fingerprinting

Maps observable signatures to app releases so facts pin to the exact version a tenant runs.

Neutral trust layer

A vendor-independent provenance schema agent runtimes read instead of re-discovering apps.

Computer-UseVerificationKnowledge