USE-CASES
What the primitives are for. One shipping flagship, and the recipes waiting for a builder — each maps directly onto primitives you can run today.
An on-chain kill switch for AI agents with wallet access. Agents spend under scoped, verifiable delegations - and that authority is revocable on a public chain (cheqd testnet), where every verifier reads the same refusal.
Built solo in a weekend at DEF CON 34 - 2nd place in the Cryptocurrency Village hackathon. The agent was Claude Desktop; a local gateway wallet held the keys, so the LLM never touched key material. A hardware kill switch flipped a StatusList2021 bit in a did:cheqd DID-Linked Resource, and the agent's next transaction was refused in 828ms - measured, not asserted (the repo's own elapsedMs). Funds never moved.
The revocation path was upstreamed into the protocol itself (cheqd DID-Linked Resources, v1.14.0), and the actual DEF CON stage credential ships in the repo, now expired - fail-closed has layers. Accountability is not a theory here; it is a circuit breaker.
or copy manually
Show me KYA-OS in action: clone https://github.com/decentralized-identity/kya-os-mcp and run the examples/revoked demo (the on-chain revocation kill switch), then explain how the per-request proof, delegation chain, and status list interact, and scaffold a minimal verifier for my stack using @kya-os/mcp.
Recipes
Patterns the primitives were designed for. No listed example yet — each one is a chance to be the first.
Human consent-gating before the tool ever runs: an ungated call gets a signed needs_authorization challenge, approval mints a scoped credential, and the server enforces it pre-execution. Consent is signed, not assumed.
Budget-bound capability tokens: a principal delegates a spend ceiling, the chain attenuates, nothing rounds up.
Agent registries (A2A, MCP, NANDA) that carry proof posture in their listings — discovery you can actually trust.
Every action leaves a trace in a tamper-evident RFC 9162 Merkle ledger - inclusion and consistency proofs, not log files.
Maintain one Entity Card; project it onto A2A, MCP, and NANDA from the same code path. Update once, consistent everywhere.
Fail-closed revocation for cards and delegations — unreachable or malformed status lists count as revoked, on-chain optional.