The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill configuration in assets/mcp.json downloads the @masonator/coolify-mcp package from the NPM registry at runtime using npx.
[REMOTE_CODE_EXECUTION]: The skill executes the @masonator/coolify-mcp package as a project-level MCP server, running third-party code within the agent's environment.
[COMMAND_EXECUTION]: The skill executes a bundled local script scripts/scaffold.js using Node.js to generate the deployment repository structure, render templates, and perform git operations.
[COMMAND_EXECUTION]: The skill utilizes bash shell built-ins (/dev/tcp probes) to perform network connectivity checks against the public IP of the target host to verify firewall rules.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external tool outputs, such as container logs and command results. While this is a vulnerability surface, the skill documentation in assets/docs/platform.md includes explicit safety guidance, advising the agent to treat log and build output as untrusted and framed by nonce-bounded boundaries.
Ingestion points: MCP logs (logs), command execution output (run_once), and infrastructure metadata (get_infrastructure_overview) in SKILL.md and assets/docs/platform.md.
Boundary markers: The skill mentions that logs and build output arrive framed as untrusted data with nonce-bounded boundaries.
Capability inventory: The skill can execute arbitrary commands in containers (run_once), write files to the local filesystem (scaffold.js), and perform network probes.
Sanitization: The skill provides explicit instructions to the agent to treat input as data and ignore instructions planted in logs.

setup