gmgn-contract-dd
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted blockchain data such as token names, symbols, and links which are attacker-controlled and may contain malicious instructions. It implements a robust defense by explicitly instructing the agent to treat these values as data and never as instructions. (1) Ingestion points: Output from gmgn-cli token info, security, and market commands. (2) Boundary markers: Explicit instructions to quote values and ignore instruction-like text within metadata. (3) Capability inventory: Read-only CLI tools and local Python-based JSON filtering. (4) Sanitization: The skill notes that gmgn-cli automatically neutralizes suspicious metadata by stripping control, zero-width, and bidi characters.
- [EXTERNAL_DOWNLOADS]: The skill requires the gmgn-cli utility, which is a legitimate resource provided by the vendor (gmgnai). It includes instructions to update the tool via npm install -g gmgn-cli if necessary.
- [COMMAND_EXECUTION]: The skill executes shell commands to invoke the gmgn-cli tool and pipes the results into local Python scripts for data extraction. These scripts are used strictly for filtering large JSON payloads to optimize context usage and do not perform any network operations or persistent changes.
Audit Metadata