LitClinic

Demo

About

LitClinic is an existing Web3 healthcare product exploring how verifiable onchain context and explicit authorization boundaries can support safer agentic workflows. For ETHOnline 2026, we built a new agentic trust and onchain context layer that extends the existing LitClinic experience. The new layer gives agents sanitized context, verifies live onchain activity, applies deterministic fail-closed policies, and introduces human-backed authorization before privileged workflows are allowed to continue. Instead of blindly trusting client input or allowing an agent to act when evidence is incomplete, LitClinic can require stronger approval or block execution entirely. A BLOCKED result is therefore an intentional security outcome, not a failure. The hackathon extension integrates The Graph for live onchain context and World AgentKit / AgentBook for agent identity and authorization checks, while keeping sensitive clinical information outside public blockchain state. The existing LitClinic UI predates ETHOnline 2026; the agentic trust and onchain context infrastructure shown in this submission was built during the hackathon.

How it's made

The ETHOnline extension is built as a TypeScript/Bun workspace with explicit boundaries between context acquisition, deterministic reasoning, and authorization. The Graph integration queries live Uniswap V3 Ethereum data and converts it into constrained onchain context, including freshness and observed wallet activity. This prevents the agent from relying only on arbitrary client-provided claims. A deterministic agent-core evaluates that evidence using a fail-closed policy. If context is missing, stale, inactive, or requires stronger authorization, the workflow returns require_approval / BLOCKED rather than allowing the agent to continue. For the authorization layer, we integrated World AgentKit and AgentBook. The live AgentBook lookup resolves the agent's registration and human-backed authorization state and feeds that result into the final execution-permission decision. We also implemented the World ID Sandbox Selfie Check workflow. Android end-to-end verification is currently pending external Google Play Internal Test access approval, so we do not claim completed Sandbox E2E in this submission. The repository also includes a typed SDK/API layer, sanitized context schemas, validation, secret scanning, deterministic demo modes, and 53 passing tests. Sensitive clinical data is deliberately kept outside public blockchain state; onchain infrastructure is used for verifiable context and authorization rather than exposing patient information.