package-netbird-green

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from configuration files to perform infrastructure management tasks, creating a vulnerability surface where malicious instructions could potentially influence agent actions.
  • Ingestion points: The colors.yml file and COLORS_PAR_* environment variables (such as COLORS_PAR_NETBIRD_HOST) are ingested to determine the deployment state.
  • Boundary markers: No explicit prompt boundary markers or "ignore instructions" directives were found in the provided files to prevent the agent from obeying instructions embedded in configuration data.
  • Capability inventory: The skill performs extensive operations including command execution (OpenTofu, Ansible), file system writes (rendering the work directory), and network API calls to Vultr and Cloudflare.
  • Sanitization: Input validation logic is referenced in the launcher as being part of the netbird library, but the validation implementation is not present in the analyzed skill files.
  • [DYNAMIC_EXECUTION]: The green launcher script dynamically loads and executes Clojure code from remote repositories during startup.
  • Evidence: The script uses babashka.deps/add-deps to resolve and pull the netbird, green, and once libraries from GitHub.
  • Evidence: It uses requiring-resolve to dynamically call and execute functions from these externally loaded namespaces.
  • [EXTERNAL_DOWNLOADS]: The launcher script initiates connections to external Git repositories to fetch its core functionality.
  • Evidence: Fetches resources from https://github.com/getcolors/netbird.git, https://github.com/getcolors/green.git, and https://github.com/getcolors/once.git. These are vendor-owned resources from the skill author.
  • [REMOTE_CODE_EXECUTION]: The skill executes code retrieved from remote sources during the initialization phase of the green launcher.
  • Evidence: The dynamic resolution and loading of the netbird library via GitHub URLs allows the script to execute remote code.
  • [COMMAND_EXECUTION]: The skill is designed to invoke multiple system commands and tools for infrastructure provisioning and management.
  • Evidence: The documentation and scripts mention the execution of OpenTofu, Ansible, Traefik, and netbird-server commands to converge the infrastructure state.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 01:39 PM