harness

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow reads Harness-generated session/objective and artifact metadata from GET /api/external-agent/owner/sessions and GET /api/external-agent/owner/sessions/{sessionId}/artifacts, which explicitly come from the Harness collaboration output (untrusted data that could include outsider-authored free text).

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill explicitly provides crypto/financial actions: it instructs the agent to sign an EVM pairing message (POST /wallet/sign), perform on-chain transfers (deposits to the provisioned wallet and payments to a quoted payTo address on Base), create withdrawal requests, settle credit purchases by submitting txHash, and call management endpoints (POST /wallet/revoke, POST /credits). These are specific wallet/crypto payment operations (token contract addresses, chains, exact transfer amounts, and tx hashes), i.e. direct financial execution capability, even though user confirmation is required.