package-langfuse-red

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The launcher script red downloads software packages from the author's GitHub repositories and the NPM registry during its first-run initialization process.
  • Evidence: The PINS constant in the red script defines specific GitHub URLs for package-langfuse-red, package-neon-red, package-once-red, and red.
  • [REMOTE_CODE_EXECUTION]: The skill implements a bootstrapping mechanism that fetches code from remote GitHub repositories and executes it using the Bun runtime. This is the primary intended function of the launcher.
  • Evidence: The installToCache function in the red script executes bun install to fetch the dependencies defined in the PINS object.
  • [DYNAMIC_EXECUTION]: The script performs runtime installation and dynamic loading of its core logic based on content-hashed cache keys.
  • Evidence: The script uses Bun.spawnSync to install dependencies into a cache directory and subsequently loads them using await import() on computed paths.
  • [COMMAND_EXECUTION]: The script executes the Bun package manager to manage dependencies and is designed to orchestrate infrastructure commands through OpenTofu and Ansible.
  • Evidence: Direct execution of process.execPath (the Bun binary) to perform installation tasks.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests infrastructure configurations from colors.yml, creating a surface for indirect prompt injection if the configuration file is derived from untrusted sources.
  • Ingestion points: The skill reads colors.yml from the working directory to define the desired state of the deployment.
  • Boundary markers: The documentation specifies a validation phase (resulting in exit code 2) that verifies the configuration and lists problems before any provider calls are made.
  • Capability inventory: The skill manages VPCs, firewalls, instances, and databases across AWS and Vultr using high-capability tools like OpenTofu and Ansible.
  • Sanitization: The script includes logic to ensure that storage, backup, and state credentials are not unsafely shared unless explicitly acknowledged by the operator in the configuration file.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 01:28 PM