sati-sdk

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

• Potentially malicious external URL detected (high risk: 0.90). The docs show the SDK/CLI will make runtime requests to MCP endpoints (e.g. "https://mcp.example.com" and example MCP endpoints like "https://myagent.com/mcp") to auto-fetch tools, prompts and resources that become agent prompts/metadata—i.e., external content fetched at runtime can directly control prompts.

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

• Direct money access detected (high risk: 1.00). The skill SDK exposes explicit on-chain transaction and wallet operations (registerAgent, giveFeedback, transfer, updateReputationScore, prepare/submit feedback flows) and requires a payer/keypair or wallet signature. These are concrete crypto/blockchain execution primitives (wallet signing, submitting transactions, paying SOL) — not generic tooling — so it grants direct crypto financial execution capability.