The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted infrastructure configuration files such as Kubernetes manifests and Terraform files, then incorporates the resulting tool outputs and violation messages into the agent context. This creates a surface for indirect prompt injection if an attacker crafts manifests designed to generate malicious violation messages that influence the agent's behavior. 1. Ingestion points: scripts/process.py (scans directories for config files), scripts/agent.py (reads policy and input data). 2. Boundary markers: Absent. 3. Capability inventory: Command execution via subprocess.run for opa, conftest, and kubectl in scripts/agent.py and scripts/process.py; report file writing. 4. Sanitization: Absent.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute standard security CLI tools including opa, conftest, and kubectl. Commands are constructed as lists of strings rather than single strings executed in a shell, which minimizes the risk of command injection.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download conftest from GitHub and opa from the official project website. These are well-known and reputable sources within the cybersecurity and DevOps domains.
[METADATA_POISONING]: There is a minor inconsistency in author metadata between the skill frontmatter (mahipal) and the license file (mukul975), which may cause confusion regarding the skill's origin.

implementing-policy-as-code-with-open-policy-agent