citation-check

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/call.mjs, the agent takes outsider-provided free text from --search (query) or key=value args, then reads the resulting JSON from GET /catalog/search or other GET/POST endpoints under https://fittings.sh without first selecting a specific trusted item.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches service definitions and payment/quote challenges at runtime from https://fittings.sh (e.g., the catalog URL https://fittings.sh/catalog?category=Research%20and%20reference), and that remotely served JSON directly controls which endpoints/inputs the client invokes and how it behaves, so this is a required runtime dependency that can change agent behavior.

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

• Direct money access detected (high risk: 1.00). The skill explicitly requires and describes executing paid calls settled in USDC on Base: it instructs setting a private key (FITTINGS_PRIVATE_KEY) as an environment variable, describes an off-chain signature authorizing a USDC transfer, and explains the facilitator submits the transaction (covering gas). These are specific payment instructions and a direct mechanism to move funds, not generic API or browsing capabilities.