The Agent Skills Directory

[SAFE]: No security issues were identified. The skill's primary function is static analysis of configuration files, and it includes robust instructional guardrails to prevent the agent from being misled by untrusted content within those files.- [PROMPT_INJECTION]: The skill incorporates a 'Prompt Injection Safety Notice' that explicitly instructs the agent to treat external content (such as GCP configuration files) as data rather than instructions. Evidence chain for potential indirect injection: (1) Ingestion points: Locates and reads Terraform, YAML, and JSON files via Glob and Read tools. (2) Boundary markers: Contains a dedicated security section warning the agent to ignore directives embedded in target files. (3) Capability inventory: Static file analysis and reporting using Read, Grep, and Glob; no network or shell execution capabilities are utilized. (4) Sanitization: Instructions clearly define that findings must be derived from technical configuration and CIS requirements, not claims made within the files.- [COMMAND_EXECUTION]: The usage of Grep and Glob tools is limited to searching for security-relevant patterns in static infrastructure-as-code files. There is no evidence of command injection vulnerabilities or attempts to spawn arbitrary subprocesses.

gcp-review