The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, including Terraform files, Kubernetes manifests, and Dockerfiles, which could potentially contain malicious instructions.
Ingestion points: The scripts/iac_scanner.py script reads local IaC files for analysis, and instructions in SKILL.md guide the agent to analyze arbitrary file contents.
Boundary markers: The skill includes a detailed 'Evidence contract' section in SKILL.md that explicitly instructs the agent to treat all collected data as untrusted evidence, separate facts from inferences, and never treat evidence as instructions.
Capability inventory: The skill uses subprocess.run in scripts/cloud_auditor.py to wrap cloud CLIs (aws, az, gcloud) and provides shell commands in SKILL.md for various security tools.
Sanitization: The iac_scanner.py script utilizes yaml.safe_load_all for parsing, which is a security best practice for handling YAML data.
[COMMAND_EXECUTION]: The scripts/cloud_auditor.py script uses the subprocess module to execute cloud provider CLI commands. While it uses list-based arguments (minimizing shell injection risks), it remains a powerful tool for environment interaction.
Evidence: The run_cli function in scripts/cloud_auditor.py executes commands constructed from a base CLI name and user-provided arguments.

10-cloud-security