The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches vulnerability enrichment data from well-known and trusted infrastructure, including the CISA Known Exploited Vulnerabilities (KEV) catalog and EPSS probability scores from FIRST.org and Empirical Security.
[COMMAND_EXECUTION]: Shell utilities including curl, gunzip, and jq are used to retrieve and process security metadata. These operations are limited to data acquisition and parsing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external vulnerability scanners and OpenVEX documents which could contain adversarial instructions.
Ingestion points: Processes scan results via a TriageInput interface, local OpenVEX JSON files, and outputs from third-party tools like knip or govulncheck (SKILL.md).
Boundary markers: The skill defines specific logic to validate OpenVEX justifications and enforces a mandatory 'non-waivable' policy for KEV entries to prevent unauthorized suppression of critical findings.
Capability inventory: Utilizes network access via curl and file processing via jq to manage and prioritize findings.
Sanitization: Implements status validation logic to reject invalid VEX assertions and uses structured parsing for external feeds.

cve-exploitability-triage