package-langfuse-green

Warn

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The green launcher script is designed to dynamically download and execute Clojure-based dependencies from remote GitHub repositories (github.com/getcolors/langfuse.git and github.com/getcolors/green.git) using the babashka.deps/add-deps function at runtime. While these resources are owned by the skill's author, the mechanism allows for the execution of remote code during the skill's operation.
  • [DYNAMIC_EXECUTION]: The script uses dynamic resolution and loading techniques, such as requiring-resolve and require, to invoke functionality from the remote libraries after they are fetched. This allows the skill to execute code that is not present in the local file system until the moment of execution.
  • [COMMAND_EXECUTION]: The skill instructions and scripts involve the execution of various system commands including npx, chmod, OpenTofu, and Ansible. These tools are used to provision virtual private clouds, firewall groups, and instances, as well as to manage DNS records and storage buckets.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes desired state configuration from the colors.yml file, which orchestrates complex infrastructure tasks across six remote machines. This creates a vulnerability surface where malicious or malformed configuration could lead to unintended deployment behaviors.
  • Ingestion points: Configuration values are read from colors.yml and environment variables prefixed with COLORS_PAR_*.
  • Boundary markers: The script does not define explicit delimiters or instructions to ignore embedded commands within the configuration data.
  • Capability inventory: The skill has the capability to write to the local file system (.colors/ directory), perform network operations with cloud providers (Vultr, Cloudflare, R2), and execute commands on remote hosts via SSH.
  • Sanitization: The launcher script includes basic validation logic that terminates execution with an exit code of 2 if errors are detected, although the comprehensive validation logic is contained within the remotely loaded library.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 12, 2026, 01:48 PM