The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches threat intelligence and CVE updates from trusted organizations and well-known services (such as CISA, NIST, MITRE, and major security vendors like Microsoft and Google). These references are managed through a hardcoded allowlist in assets/trusted_sources.json to prevent the ingestion of untrusted content.
[COMMAND_EXECUTION]: The agent is instructed to execute a local Python script (scripts/check_source.py) to validate candidate URLs against the allowlist before any network operations occur. This provides a deterministic security gate that cannot be bypassed by the LLM's own interpretation of the domains.
[PROMPT_INJECTION]: The skill explicitly addresses the risk of indirect prompt injection by directing the agent to treat all fetched content as data rather than instructions. It includes a specific protocol to identify and report embedded directives (e.g., 'ignore previous instructions') while refusing to execute them.

cybersecurity-expert