The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of code diffs provided by users. This represents a potential attack surface where malicious instructions could be embedded within the code being reviewed. However, the risk is mitigated by the skill's instructions which enforce a highly structured and limited output format (one-line findings with specific tags), effectively constraining the agent's response to a predefined template and reducing the likelihood of executing instructions found in the analyzed content.

ponytail-review