The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python validation scripts (check_review.py, check_override.py, jev_gate.py) used to verify the semantic consistency of product review artifacts. These scripts perform local file reads and JSON parsing with safety checks (such as size limits and duplicate key detection) but do not execute arbitrary shell commands or perform network operations.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data such as product briefs, evidence receipts, and external feedback. It significantly mitigates injection risks by using a strict 'Trust Chain' model where all external data is treated as untrusted and must be validated against JSON schemas. It further utilizes a 'Human Judgment Gate' for sensitive or high-risk decisions, ensuring that the AI agent does not autonomously act on potentially poisoned instructions embedded in processed data.
[SAFE]: The skill follows security best practices by explicitly stating that it does not automatically install dependencies or configure remote providers. It provides a requirements-dev.txt for the standard jsonschema library, which is a common and safe practice for local development and validation.

product-design-harness