Agent Trust Stack

An interactive golden path for trustworthy AI-agent actions. Review a candidate skill before installation, enforce a deterministic payment policy before signing, and export a SHA-256-linked Action Receipt after an allowed or human-approved decision.

Live demo

https://feeeeelixwong.github.io/agent-trust-stack-demo/

What the flow proves

Gate	Demonstrated behavior	Production layer
Inspect capability	A risky, unpinned skill is blocked before it can receive authority.	Audit Skill Supply Chain
Authorize spend	An allowlist, automatic threshold, per-action limit, and human-review path produce deterministic outcomes.	AgentSpend Guard
Preserve evidence	The evaluated intent, source hash, decision, timestamp, and previous hash are bound into an exportable receipt.	IntentProof

Run locally

npm install
npm run dev

Run the quality gates with:

npm run lint
npm test
npm run build

Safe by design

This is a browser-only dry run. It never connects a wallet, requests a secret, signs a payment, transfers funds, or represents simulated evidence as an on-chain settlement. See ARCHITECTURE.md for the trust boundaries and production replacement points.

License

MIT. See LICENSE.