package-rybbit-green

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The green launcher script downloads Clojure dependencies from the author's GitHub repositories (github.com/getcolors/rybbit, github.com/getcolors/green, github.com/getcolors/once) at runtime. The library dependencies use pinned commit SHAs (e.g., 26e6f83cc28d74a6aff9d93a3b8001cc894a1883) to ensure code integrity and prevent unauthorized changes.
  • [COMMAND_EXECUTION]: The skill executes lifecycle commands such as build, create, and delete to provision cloud resources on providers like Vultr or DigitalOcean. It also manages the ~/.ssh/config file and deployment keys to facilitate secure access to the provisioned infrastructure.
  • [DATA_EXFILTRATION]: The skill manages sensitive infrastructure credentials, including Cloudflare API tokens, object storage access keys, and compute provider API keys. These are handled via environment variables and are required for its primary function of resource provisioning and automated data backups.
  • [DYNAMIC_EXECUTION]: The launcher script employs Babashka's add-deps to fetch code and requiring-resolve to dynamically load and execute the CLI and workflow logic from its resolved library dependencies.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill reads desired state and topology configuration from a local colors.yml file.
  • Boundary markers: Documentation states that validation is handled by the core library, which is designed to fail closed in the event of unreadable state or provider mismatches.
  • Capability inventory: Remote resource provisioning, SSH key management, network policy configuration, and automated R2/S3 backups.
  • Sanitization: Includes specific checks to prevent environment variable overrides (such as COLORS_PAR_PROFILE) from redirecting the skill to unintended project states or external resource definitions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 01:53 PM