The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze external, potentially untrusted content such as security advisories, patches, and research papers, creating an attack surface for indirect prompt injection.
Ingestion points: Remote evidence collection is described in SKILL.md and references/retrieval-and-provenance.md.
Boundary markers: The instructions explicitly direct the agent to treat source content as untrusted data and to load specific hygiene behaviors to prevent redirection of actions.
Capability inventory: The skill itself contains no executable code or scripts, relying instead on the agent's built-in tools for data retrieval.
Sanitization: The instructions require the agent to distinguish between source claims and local observations and strictly prohibit the execution of downloaded artifacts without a separate review workflow.
[EXTERNAL_DOWNLOADS]: The skill references fetching data from well-known external security services.
Evidence: references/sources.md lists official resources including the CVE Program, NIST NVD, CISA KEV, and MITRE ATT&CK.
Context: These are authoritative technology services and government repositories. The skill provides clear instructions against executing any content downloaded from these sources during the research phase.

deep-research-offensive