Skip to content
Flowmark

Design with AI agents

Flowmark gives agents an expert diagram-design workflow and deterministic feedback instead of asking them to memorize a changing grammar or judge an SVG from intuition alone.

Install the official skill

Section titled “Install the official skill”

The open Skills CLI currently discovers skills from Git sources. Install Flowmark’s skill directly from its package directory:

Terminal window
npx skills add https://github.com/ministryplace/flowmark/tree/main/packages/flowmark-agents

The same files ship in @ministryplace/flowmark-agents for dependency-managed agent environments. The skill is host-neutral and can be installed into Codex, Claude Code, Cursor, and other clients supported by the Skills CLI.

What the skill teaches

Section titled “What the skill teaches”

The design-flowmark-diagrams skill guides an agent to:

  1. Identify the audience, decision, and one question the diagram must answer.
  2. Separate supplied facts from inference instead of inventing architecture.
  3. Choose an appropriate architecture, event, workflow, ERD, or sequence pattern.
  4. Keep Flowmark source as the durable source of truth.
  5. Validate, format, render, inspect, and repair using the Flowmark CLI.
  6. Return source, artifacts, assumptions, and unresolved questions together.

Detailed pattern, repair-order, and delivery references load only when needed, keeping the agent’s working context compact.

Use it with a repository

Section titled “Use it with a repository”

Give the agent a precise reader question and identify which repository facts it may inspect:

Use the Flowmark diagram skill to show a new engineer how checkout moves from the web app through payment authorization. Read the relevant code and configuration, distinguish observed facts from assumptions, and return the Flowmark source plus a portable SVG.

Review the assumptions as carefully as the diagram. Generated source should stay in version control beside the system it explains, and CI can verify it with flowmark check and flowmark render.

The agent-native direction

Section titled “The agent-native direction”

The skill is the first layer. Flowmark’s machine-readable capability manifest, deterministic quality diagnostics, semantic comparison, and MCP adapter will expose the same underlying public operations without duplicating parsing or rendering behavior across agent hosts.