gmgn-dev-score

Warn

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script dev_score.py uses subprocess.run to execute the gmgn-cli binary. While this is the intended functionality of the skill, it passes user-supplied arguments (chain, dev_address) directly to the shell command. Although some sanitization is implemented via the safe() function for display, the initial command execution relies on the integrity of the gmgn-cli tool and shell-level handling.
  • [EXTERNAL_DOWNLOADS]: The skill performs multiple automated network requests to gmgn.ai APIs via the gmgn-cli tool to fetch wallet history, token info, and trading stats. It also fetches data from ipv6.icanhazip.com to check the user's network configuration.
  • [DYNAMIC_EXECUTION]: The SKILL.md file contains instructions for the agent to execute shell commands (python3 ~/.claude/skills/gmgn-dev-score/dev_score.py ...) and use the output to generate a report. It also includes conditional logic that dictates agent behavior based on the command's exit code or specific JSON outputs (e.g., scored: false).
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from a highly adversarial environment (blockchain token metadata, symbols, and wallet addresses). The developer has acknowledged this by implementing a safe() function to strip ANSI escapes and non-printable characters, but the risk remains that malicious data from the API could influence the agent's report generation or logic.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 3, 2026, 11:23 AM
Security Audit — agent-trust-hub — gmgn-dev-score