gmgn-cooking

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the gmgn-cli utility to execute blockchain transactions and query statistics. It also invokes system tools like ifconfig, ip, and base64 for environment configuration and processing image data.
  • [DATA_EXFILTRATION]: The skill instructs the agent to read local file paths provided by the user and convert their contents to Base64 to be used as token logos. This creates a surface for potential data exposure if an attacker provides paths to sensitive local files (such as SSH keys or environment variables) instead of image files.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input for token names, symbols, and descriptions to populate command arguments. While it includes warnings that the backend filters for injection patterns, the lack of local sanitization or strict boundary markers poses a risk of argument injection.
  • Ingestion points: Local file paths for the --image flag and user-supplied strings for token metadata fields in SKILL.md.
  • Boundary markers: Absent; the agent is not instructed to use specific delimiters or escape sequences when interpolating user data into commands.
  • Capability inventory: The agent can execute irreversible financial transactions, spend funds, read local files, and perform network requests via the CLI tool.
  • Sanitization: The documentation relies on the CLI or backend API to perform sanitization against control characters and injection framing.
  • [EXTERNAL_DOWNLOADS]: The skill directs the agent to ensure the gmgn-cli package is installed via npm. This package is maintained by the skill's author, gmgnai, and is a legitimate dependency for the intended functionality.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 09:16 PM