package-once-green

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The green launcher script fetches dependencies from the vendor's official repositories on GitHub (github.com/getcolors/once.git and github.com/getcolors/green.git) using specific commit SHAs for version pinning.
  • [DYNAMIC_EXECUTION]: The launcher uses Babashka to dynamically load and resolve Clojure namespaces from the fetched libraries at runtime to execute deployment workflows.
  • [COMMAND_EXECUTION]: The skill instructions involve running the ./green CLI tool to perform various deployment operations including provisioning, builds, and dry-runs.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided configuration in colors.yml and environment variables to perform deployment tasks.
  • Ingestion points: Ingests configuration from colors.yml and environment variables (documented in SKILL.md and green).
  • Boundary markers: Explicit instructions are provided to the agent to avoid reading private files (such as .envrc.private) and to only request variable names rather than their secret values.
  • Capability inventory: The skill manages SSH keys, performs network API calls to cloud providers (DigitalOcean, Cloudflare, Resend, GitHub), writes to a .colors/ directory, and executes shell commands via the green launcher.
  • Sanitization: The skill utilizes validation steps through ./green build and dry-run execution prior to convergence.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 04:47 AM