The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's security audit feature acts as an ingestion point for untrusted user-provided code, creating a surface for indirect prompt injection. Malicious instructions could be embedded in code comments to manipulate the agent's output. The skill lacks explicit boundary markers for this untrusted input while maintaining capabilities to generate shell commands and new source files.
[DYNAMIC_EXECUTION]: The skill generates executable Solidity contracts and JavaScript deployment scripts based on user-supplied parameters. While it utilizes trusted templates and security libraries like OpenZeppelin, the generation of executable files from user requirements is a form of dynamic code construction that requires manual verification by the end user.

blockchain-dev