​
AI Coding Agent Skills

When using AI coding assistants (such as Google Antigravity, Gemini CLI, Claude Code, or Cursor), you can install the official dmint-policy-manager skill. This equips your AI coding agent with the system prompts, threat models, and authoring guidelines required to create and verify Dmint policies directly in your chat interface.

​
Installing the Skill (dmint install-skill)

The dmint-cli package bundles the official skill and provides an installation command:

​
Non-Interactive & Automated Installation

For scripted setups or Docker environments:

​
What the Skill Teaches Your AI Agent

Once installed in your project, your AI assistant automatically adheres to these invariants:
  1. Least Privilege Principles: When drafting policies, the agent defaults unmentioned actions to DENY.
  2. Wildcard Containment: The agent is instructed never to replace specific resources with broad wildcards (*) without explicit developer confirmation.
  3. Verification Before Deployment: The agent will run dmint verify-policy to ensure syntactic and semantic validity before committing policy files.
  4. Separation of Concerns: The agent understands that an LLM can draft policies at development time, but can never participate in runtime authorization.

​
Security Hardening

The dmint install-skill command includes strict path sanitization:
  • Whitelists skill names against ^[a-zA-Z0-9_-]+$.
  • Rejects path traversal payloads (../, ..\\, absolute root escapes).
  • Uses atomic file writes with file descriptor validation to prevent symlink hijacking.