The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user input, such as product names, categories, and persona cards, which are interpolated into prompts to generate structured search query bundles. This creates an attack surface for indirect prompt injection where malicious data could attempt to override agent instructions.
Ingestion points: As described in workflow.md, the skill accepts untrusted data through the category, product_name, and persona_card inputs.
Boundary markers: The skill lacks explicit boundary markers or delimiters for user-provided strings within its prompt structure.
Capability inventory: The skill is primarily generative and does not utilize high-privilege tools or execute external code during normal operation.
Sanitization: The skill employs a detailed post-generation validation process using schema.json and scripts/validate.js to enforce specific formatting and content constraints, mitigating accidental or malicious formatting deviations.
[SAFE]: The skill includes a local utility script scripts/validate.js and associated configuration files (package.json, package-lock.json) for output verification. These components use the well-known ajv library for JSON Schema enforcement and do not perform any unauthorized network calls, data exfiltration, or privileged operations.

100x-search-query