The Agent Skills Directory

[COMMAND_EXECUTION]: The analyze.py script executes system commands via subprocess.run(). The implementation is secure, using a list-based argument structure rather than string interpolation and specifically avoiding shell=True, which prevents shell-level command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (blockchain metadata) that may contain user-controllable strings like wallet names or Twitter handles. Because the agent is instructed to paste this data verbatim, it could theoretically be used to relay malicious instructions to the agent.
Ingestion points: analyze.py fetches holder metadata, names, and tags via the gmgn-cli tool.
Boundary markers: Absent. The SKILL.md file instructs the agent to paste the script output verbatim without employing delimiters or specific safety instructions to disregard embedded commands.
Capability inventory: The script performs subprocess calls and processes external data to generate reports.
Sanitization: Absent. The script prints raw data fields directly to the standard output as received from the API.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install gmgn-cli via npm. As this tool is provided by the skill's author (gmgnai), it is a vendor-owned resource and is categorized as safe functionality.

gmgn-holder-analysis