The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from domain research tools, which creates a surface for indirect prompt injection where malicious instructions could be embedded in records such as WHOIS or DNS responses.
Ingestion points: Data retrieved via the research tools listed in SKILL.md and references/sandbase-api-map.md, including strale_whois_lookup, strale_dns_lookup, strale_ssl_check, strale_domain_age_check, strale_domain_reputation, and strale_header_security_check.
Boundary markers: The skill includes guidelines to separate factual observations from analysis and cite sources, which serves as a basic instruction-level boundary.
Capability inventory: The skill is limited to research tool calls through the SandBase gateway and does not have access to file system writes, arbitrary shell execution, or independent network operations.
Sanitization: No explicit technical sanitization, filtering, or schema validation for the content of the external records is described.

domain-intelligence