@vohlsyr/plugin-trustfetch

ElizaOS plugin for TrustFetch -- prompt-injection scanning and clean webpage fetching for AI agents that have to read untrusted content.

Paid per call via x402 -- USDC on Base, straight from the agent's own wallet. No API keys, no signup.

Actions

SAFE_FETCH_URL

Fetches a URL as clean Markdown AND scans it for prompt-injection attempts in one call. ~$0.03/call.

"can you safely read https://example.com/article for me?"

SCAN_FOR_INJECTION

Scans a piece of text you already have (not a URL) for hidden prompt-injection attempts. ~$0.02/call.

"scan this text for injection: '...'"

Setup

npm install @vohlsyr/plugin-trustfetch

Add to your character/project's plugin list, and set one secret:

EVM_PRIVATE_KEY=0x...   # a wallet funded with a small amount of USDC on Base

This is the same setting name ElizaOS's own @elizaos/plugin-evm uses, so an agent that already has a funded wallet configured for other on-chain actions picks it up automatically -- no second key to manage.

How it works

1. The action calls the real endpoint (e.g. POST /tools/safe-fetch).
2. The server responds 402 Payment Required with the price and payment details.
3. x402-fetch signs and attaches a payment using the agent's wallet, and retries the request.
4. The server settles the payment and returns the real result.

Development

npm install
npm run typecheck
npm test        # vitest -- mocks the x402-fetch boundary, exercises real
                 # URL-extraction/request-building/formatting logic
npm run build

The plugin also bundles a zero-cost elizaos test case (trustfetch plugin sanity) that pings the service's free /healthz endpoint -- no wallet or payment required for that one.

License

MIT