erc-8004

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on the bankr CLI and jq for core operations.
  • Executes bankr prompt to facilitate ETH bridging from Base to Mainnet and to submit registration transactions to Ethereum.
  • Uses jq to parse JSON responses from RPC providers and Pinata.
  • [EXTERNAL_DOWNLOADS]: Fetches external data from the Ethereum blockchain and decentralized storage.
  • Queries Ethereum JSON-RPC endpoints (Alchemy, LlamaRPC) to retrieve on-chain agent metadata.
  • Downloads agent registration files from IPFS via Pinata and ipfs.io gateways, or from arbitrary HTTP URLs provided in the registry.
  • [DATA_EXFILTRATION]: Uploads agent registration data to Pinata Cloud.
  • The upload-to-ipfs.sh script sends JSON profile data to api.pinata.cloud using a user-supplied PINATA_JWT for authentication.
  • [REMOTE_CODE_EXECUTION]: Utilizes node -e for dynamic data processing.
  • Executes short Node.js snippets to encode Ethereum ABI calldata and decode hex strings returned from contract calls.
  • [PROMPT_INJECTION]: Vulnerable to indirect prompt injection through external profile ingestion.
  • Ingestion points: scripts/get-agent.sh fetches agent profiles from attacker-controlled or third-party IPFS CIDs and HTTP URLs.
  • Boundary markers: Absent; the fetched profile content is printed directly to the console or processed as JSON without instruction-filtering.
  • Capability inventory: The agent has the ability to move funds (bridge ETH) and modify identity registries via the bankr tool.
  • Sanitization: Profiles are formatted via jq for display, but no sanitization of natural language instructions within the metadata is performed.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 06:58 AM