skills/cascade-protocol/sati/sati-sdk/Gen Agent Trust Hub

sati-sdk

Fail

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [CREDENTIALS_UNSAFE]: A hardcoded Base58-encoded private key was detected in the file packages/sdk/scripts/register-echo-agent.ts. While the script appears to be a developer utility for a demo agent, hardcoding sensitive credentials poses a significant risk if the script is used in a production context or if the key is not strictly limited to a test environment.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through the ingestion of untrusted agent metadata.
  • Ingestion points: The function fetchRegistrationFile in packages/sdk/src/registration.ts fetches JSON metadata from external URIs (IPFS, Arweave, or HTTP).
  • Boundary markers: The skill uses a Zod schema (RegistrationFileSchema) to validate the structure of the fetched data.
  • Capability inventory: The skill can perform file system operations (via scripts), upload to IPFS, and sign Solana transactions for agent registration and feedback.
  • Sanitization: The metadata is sanitized using safeParse through the Zod schema before being processed.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to several well-known and vendor-specific services:
  • Fetches market data from api.elections.kalshi.com for AI predictions.
  • Downloads agent metadata from IPFS (ipfs.io) and Arweave (arweave.net).
  • Connects to Solana RPC nodes via Helius (helius-rpc.com) and vendor proxies (sati.cascade.fyi).
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 7, 2026, 10:20 PM