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.ymlfile andCOLORS_PAR_*environment variables (such asCOLORS_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
netbirdlibrary, but the validation implementation is not present in the analyzed skill files. - [DYNAMIC_EXECUTION]: The
greenlauncher script dynamically loads and executes Clojure code from remote repositories during startup. - Evidence: The script uses
babashka.deps/add-depsto resolve and pull thenetbird,green, andoncelibraries from GitHub. - Evidence: It uses
requiring-resolveto 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, andhttps://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
greenlauncher. - Evidence: The dynamic resolution and loading of the
netbirdlibrary 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, andnetbird-servercommands to converge the infrastructure state.
Audit Metadata