The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The Python scripts (agent.py, process.py) fetch vulnerability catalogs and exploit prediction scores from trusted organizations, including CISA (cisa.gov), FIRST.org (api.first.org), and NIST (nist.gov). These network operations are necessary for the skill's primary function and target well-known, reputable domains.
[REMOTE_CODE_EXECUTION]: Documentation files (SKILL.md and workflows.md) contain examples of piping data from a remote URL to a Python interpreter. While this pattern is generally higher-risk, the source in these examples (cisa.gov) is a trusted government organization, and the command is provided as a legitimate example for processing public vulnerability feeds.
[DATA_EXFILTRATION]: The skill transmits CVE identifiers to external APIs to retrieve context for vulnerability triage. This behavior is documented and expected for a vulnerability management tool; it does not involve the collection or transfer of sensitive user credentials or private data.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by ingesting untrusted vulnerability scan results from CSV and XML files.
Ingestion points: scripts/process.py (CSV/XML ingestion)
Boundary markers: Absent
Capability inventory: scripts/process.py and scripts/agent.py (File system write, Network GET requests)
Sanitization: Standard parsing is performed using the csv and xml.etree.ElementTree libraries

triaging-vulnerabilities-with-ssvc-framework