The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py uses subprocess.run to execute gcloud CLI commands. This is the primary mechanism for auditing the GCP environment. The implementation follows best practices by using argument lists instead of shell strings, which prevents shell injection vulnerabilities.- [DATA_EXPOSURE]: The agent collects GCP infrastructure metadata, including IAP IAM policies and Access Context Manager levels. This sensitive configuration data is used solely to generate a local JSON audit report and is not transmitted to any external third-party domains.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of system command outputs from gcloud. It identifies ingestion points in scripts/agent.py where JSON results are parsed. No unsafe interpolation into LLM prompts was observed, and the capability inventory is limited to read-only audit operations.

implementing-zero-trust-with-beyondcorp