The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill implements a command import_templates_from_git which allows downloading and installing agent templates from arbitrary Git URLs. Since these templates contain Python scripts (pyscript connectors), this enables remote code execution if a user is directed to an untrusted repository.
[COMMAND_EXECUTION] (HIGH): The skill provides end-to-end workflows for creating and executing custom Python logic through connectors (create-template and install-template). It specifically facilitates the deployment of executable code to local and production environments.
[CREDENTIALS_UNSAFE] (HIGH): A core feature of the skill is managing the Machina vault (configure-secrets.md). It provides commands to create, verify, and delete secrets, including sensitive data like Google Cloud service account JSON and API keys. This high-privilege access is a significant risk if the agent is compromised.
[DATA_EXPOSURE] (MEDIUM): The SKILL.md file reveals hardcoded absolute local paths (/Users/fernando/machina/...), which exposes information about the host's directory structure and user environment.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill exhibits an attack surface for indirect injection by ingesting template data from external Git repositories and local files while possessing high-privilege capabilities (file writing, secret management, and code execution).

template-constructor