The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using python3 -c or bc to perform Wei conversions. This pattern involves spawning subprocesses with arguments (AMOUNT and DECIMALS) derived from external input, which can be risky if the values are not strictly validated.
[DYNAMIC_EXECUTION]: The use of python3 -c to evaluate a mathematical expression at runtime constitutes dynamic code execution. While intended to ensure mathematical precision for token decimals, it provides a vector for arbitrary code execution if a user-supplied amount or a retrieved decimal value contains malicious code snippets.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the KyberSwap Token API, Aggregator API, and CoinGecko. This untrusted external data is interpolated into prompts and output tables without clear boundary markers or sanitization, potentially allowing malicious content in token metadata to influence agent behavior.
Ingestion points: Data retrieved via WebFetch from aggregator-api.kyberswap.com, token-api.kyberswap.com, and api.coingecko.com.
Boundary markers: None identified for the inclusion of external API content in the agent's response context.
Capability inventory: The skill has network access, shell command execution capabilities, and local file system read access.
Sanitization: No specific sanitization or escaping is mandated for text fields like token names or exchange identifiers before they are displayed or processed.
[EXTERNAL_DOWNLOADS]: The skill fetches token information and routing data from the vendor's official APIs (kyberswap.com) and well-known services like CoinGecko.

quote