package-once-blue
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The blue launcher script uses the "uv" tool to download dependencies from the author's GitHub repositories, including github.com/getcolors/once.git, github.com/getcolors/blue.git, and github.com/getcolors/colors-compute.git.
- [REMOTE_CODE_EXECUTION]: The launcher script downloads and executes code from the specified GitHub repositories at runtime. The use of specific commit hashes for these dependencies is a positive security practice that mitigates certain supply chain risks.
- [COMMAND_EXECUTION]: The skill requires the agent to execute shell commands using the "./blue" launcher to perform tasks such as building, dry-running, and managing deployments.
- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it processes user-provided configuration from colors.yml and processes remote infrastructure state. Ingestion points: colors.yml and remote compute inventory. Boundary markers: The instructions tell the agent to ignore specific generated directories and private environment files. Capability inventory: The skill can execute commands that modify cloud infrastructure and publish credentials to GitHub. Sanitization: The skill mandates that secrets be provided via environment variables rather than being stored in configuration files.
- [DYNAMIC_EXECUTION]: The blue launcher utilizes "uv run" to dynamically resolve and execute Python scripts and their dependencies at runtime.
Audit Metadata