The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md contain strict 'HARD LIMITS' that explicitly forbid the agent from using bash to process, sum, or filter the returned API data. This is a proactive security measure to prevent command injection or unnecessary sub-process spawning.
[DATA_EXFILTRATION]: All network requests are directed to the official API domain https://api.tokenomist.ai using a proxied client. The skill requires an environment variable TOKENMIST_API_KEY, but includes logic in tools/tokenomist_tools.py (_safe_error_message) to redact sensitive keys or placeholders from being leaked in tool output or error logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from an external API. While the data is structured (JSON), the skill processes fields like allocationName and symbol. The vulnerability surface is minimized by the instructions to the agent to treat the output as data to be formatted, rather than instructions to be followed.
[SAFE]: The code follows standard patterns for API integration, using a dedicated client and well-defined tool wrappers. No signs of obfuscation, persistence, or privilege escalation were detected.

tokenomist