The Agent Skills Directory

[SAFE]: The skill consists of several Node.js scripts used to compare a website build against a baseline. The scripts perform local file system operations (reading/writing JSON and HTML files) and network requests to a user-specified base URL for live site verification.
[COMMAND_EXECUTION]: The run-all.mjs script uses child_process.spawnSync to execute other verification scripts located within the same directory. This is a standard and safe implementation for a master test runner script.
[DATA_EXPOSURE_EXFILTRATION]: The verify-live.mjs script uses the standard fetch API to perform HTTP requests against a user-provided --base URL. This is used exclusively for verifying the response status and headers of a deployed site against the baseline. No sensitive data is exfiltrated to unauthorized domains.
[INDIRECT_PROMPT_INJECTION]: While the scripts process external data (captured HTML content), they do so using standard parsing libraries (parse5) and comparison logic. They do not interpret the content as instructions for an LLM, effectively mitigating prompt injection risks in the context of their execution.

parity-gates