Overview¶

DV Flow Manager has three distinct AI-related capabilities. They serve different audiences, and this page helps you find the right one.

The three capabilities¶

Driving DFM from an external agent.: You use an external coding agent (Claude, GitHub Copilot, and others) and want it to discover, understand, and run your DFM project. The agent calls dfm commands, reads AGENTS.md, and consumes JSON discovery output. → Using DFM with External Agents
Authoring agent resources in a package.: You are building a package and want to ship resources – Skills, Personas, Tools, and References – so that any agent working with your package starts with the right context. → AI Agent Integration
Using the built-in native agent.: You want DFM’s own dfm agent TUI, which runs an LLM-driven assistant against your project with a configurable provider. → Native AI Agent (dfm agent)

If you want to…	Go to
Expose your project to Claude/Copilot/etc.	Using DFM with External Agents
Discover tasks and skills as JSON (`dfm show ... --json`)	Using DFM with External Agents
Override parameters from an agent (`-D` / `-P`)	Using DFM with External Agents
Ship Skills / Personas / Tools / References with a package	AI Agent Integration
Define a reusable skill document (`std.AgentSkill`)	AI Agent Integration
Run DFM’s built-in `dfm agent` assistant	Native AI Agent (dfm agent)

Skill – a Markdown document (and optional examples) that teaches an agent how to use a package or project. Defined with std.AgentSkill and discoverable via dfm show skills.
Persona – a behavioral profile that shapes how the agent responds.
Tool – an external tool or MCP server the agent may invoke.
Reference – supplementary files or URLs the agent should consult.

Skills, personas, tools, and references are all agent resources; see AI Agent Integration for how to author them.