The Agent Skills Directory

[PROMPT_INJECTION]: The skill features a dedicated 'Prompt Injection Safety Notice' that instructs the agent to treat all strings, comments, and descriptions in scanned files as data rather than instructions. This effectively mitigates indirect prompt injection risks by explicitly telling the agent to disregard inline claims of compliance or suppression directives (e.g., '# checkov:skip').
Ingestion points: Processes external IaC files (Terraform, CloudFormation, etc.) using the Read tool.
Boundary markers: Includes instructions for the agent to maintain strict boundaries between its rules and the content being analyzed.
Capability inventory: Access is restricted to Read, Grep, and Glob tools; it lacks network access or general shell execution capabilities.
Sanitization: Instructs the agent to ignore and report scanner suppression comments, treating them as findings rather than operational commands.
[COMMAND_EXECUTION]: The skill uses a limited toolset (Read, Grep, Glob) for static analysis. There are no patterns involving exec, eval, or subprocess spawning that would allow for arbitrary code execution.
[DATA_EXFILTRATION]: No network tools (e.g., curl, fetch) are included in the allowed-tools configuration. The skill operates exclusively on local files for auditing purposes.
[SAFE]: The skill references reputable industry standards including OWASP, SLSA v1.0, and CIS Benchmarks. The detection patterns provided for hardcoded secrets and public exposure are standard security auditing regexes and pose no risk to the environment.

iac-security