The Agent Skills Directory

[CREDENTIALS_UNSAFE] (MEDIUM): In scripts/setup.py, the validate_connection and list_projects functions explicitly disable SSL certificate verification (ssl.CERT_NONE). If a user runs this setup on an untrusted network, an attacker could intercept the x-api-key header through a Man-in-the-Middle (MitM) attack, gaining full control over the user's Dokploy instance.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill configures Claude to run @ahdev/dokploy-mcp using npx -y. This downloads and executes code from an external, untrusted npm scope at runtime. While this is the primary purpose of the skill, the lack of a pinned version or a trusted author increases the risk of a supply chain attack.
[CREDENTIALS_UNSAFE] (LOW): The setup script saves the Dokploy API key in plain text within ~/.claude/mcp.json. Although this is standard for MCP configurations, it results in sensitive credentials being stored unencrypted on the local filesystem.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill provides 67 tools that interact with a Dokploy instance. An attacker who can influence the metadata of a Dokploy project (e.g., project names or app descriptions) could potentially attempt to influence the agent's behavior when it lists or reads these resources.
Ingestion points: scripts/setup.py (via list_projects), and the @ahdev/dokploy-mcp tool outputs.
Boundary markers: None explicitly implemented in the setup script.
Capability inventory: Full control over Docker deployments, databases, and environment variables on the target server.
Sanitization: None detected in the provided Python script.

dokploy-deploy