skills/mukul975/anthropic-cybersecurity-skills/configuring-microsegmentation-for-zero-trust/Gen Agent Trust Hub
configuring-microsegmentation-for-zero-trust
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external network flow data and workload labels to generate security recommendations, which creates an attack surface where malicious data could influence security outcomes.\n
- Ingestion points: The
scripts/process.pyscript parses external CSV and JSON files provided via the--flows,--policy, and--labelscommand-line arguments.\n - Boundary markers: No delimiters or specialized instructions are present to prevent the agent from misinterpreting data values as instructions during policy generation.\n
- Capability inventory: The skill can write proposed firewall rules to the local filesystem and interact with AWS and Illumio APIs to perform security audits.\n
- Sanitization: The script uses
int()casting for ports and byte counts, but no sanitization or validation is applied to string-based labels or workload identifiers which are incorporated directly into security rules.
Audit Metadata