setup-coolify-devops

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill frequently executes local shell commands for system discovery and configuration. This includes using tailscale to check machine status, jq to parse JSON output, and curl for public IP discovery. It also performs network connectivity tests using bash's /dev/tcp device. Additionally, the bundled Node.js script scaffold.js uses spawnSync to execute git init and git commit commands.
  • [PERSISTENCE]: The skill provides instructions for the user to establish persistent environment variables by appending a source command to their ~/.bashrc file. This is done to ensure that the COOLIFY_ACCESS_TOKEN and COOLIFY_BASE_URL are available across shell sessions for the MCP server to function correctly.
  • [EXTERNAL_DOWNLOADS]: The skill interacts with several external APIs and services. It fetches the host's public IP from api.ipify.org, retrieves webhook CIDR ranges from api.github.com/meta, and performs authentication and device listing via the api.tailscale.com endpoint. These are well-known technology services and are documented as part of the skill's infrastructure management role.
  • [PRIVILEGE_ESCALATION]: The skill includes instructions to use sudo for administrative tasks such as configuring Tailscale SSH (tailscale set --ssh) and managing machine tags. These operations are restricted to the Coolify host machine and are intended for instance provisioning.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests user-supplied strings during an initial interview process and renders them directly into the agent's operating rules (CLAUDE.md) and documentation using a template engine in scaffold.js.
  • Ingestion points: User input from the Step 1 Interview is stored in instance.yaml and then rendered into other files.
  • Boundary markers: None are present in the rendering logic; the values are interpolated as raw strings.
  • Capability inventory: The skill has capabilities for file system modification, command execution (git, tailscale), and network operations.
  • Sanitization: The rendering script performs minimal unquoting of values but does not filter for instruction-like patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 05:20 PM
Security Audit — agent-trust-hub — setup-coolify-devops