The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON catalogs which could contain malicious instructions. It mitigates this risk by including explicit instructions to treat input data as untrusted and by performing data sanitization. * Ingestion points: Untrusted JSON data is ingested from the file path specified in the --input argument in scripts/build_finder.py. * Boundary markers: The SKILL.md file contains a 'Working agreement' section that explicitly instructs the agent to ignore instructions embedded in input documents. * Capability inventory: The skill performs file read/write operations via scripts/skill_io.py but lacks network capabilities or shell execution of ingested data. * Sanitization: The build_finder.py script validates product URLs to ensure they are HTTPS or root-relative and escapes JSON characters when generating the HTML prototype to prevent data injection into the HTML template.
[DYNAMIC_EXECUTION]: The Python script generates an HTML file with embedded JavaScript. This is the intended behavior for creating a rule-based prototype. The script uses secure JSON serialization and character escaping to ensure the generated file is safe to open locally.

product-finder-builder