masumi

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill includes a curl example that instructs placing an API key directly in an HTTP header (token: YOUR_API_KEY), which is an insecure pattern requiring verbatim secret insertion into commands/outputs.

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow ingests outsider-authored free text when a buyer submits input_data.dataset and analysisType through the MIP-003 /start_job request, which is then embedded into an LLM prompt for agents (e.g., LangGraph llm.invoke uses state.dataset/state.analysisType).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill includes explicit installation/runtime commands that clone and run remote code from https://github.com/masumi-network/masumi-payment-service (e.g., "git clone ...", "npm install", "npm run dev"), which fetches and executes external code during setup/runtime.

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

• Direct money access detected (high risk: 1.00). The skill is explicitly a payments system built on Cardano: it describes agent-to-agent payments, a Masumi Payment Service (wallet management, payment processing), smart-contract escrow payment flow, three-wallet architecture (purchasing/selling/collection wallets), USDM stablecoin, and example Payment Service API usage (curl to register, with fees). These are specific financial execution capabilities (wallet management, sending/collecting funds, and on-chain transactions), not merely generic tooling.