Add graph-wizard CLI workflow

[dgroomes]

Summary

• add a Bun-native graph-wizard CLI that turns a natural-language prompt into a diagram spec, maxGraph XML, mxGraph-compatible XML, PNG render, JSON report, and markdown summary
• add render-only browser mode plus diagram reporting/export helpers so generated diagrams can be rendered and inspected automatically
• refine the heuristic/revision flow so network-topology prompts wire boundary nodes into the main request path instead of leaving them isolated
• add a reusable graph-wizard agent skill under llm-agent/skills/graph-wizard/SKILL.md
• mirror the skill into .cursor/skills/graph-wizard/SKILL.md so Cursor Agent can discover it from the workspace
• document the new Bun/Bash-first workflow and keep Nushell support as a thin wrapper

Testing

• bun run build-client
• bun run check
• bun graph-wizard.ts "please give me a graph of a network topology for how AWS Lambda functions work and please show the logging infrastructure" --provider heuristic --max-iterations 2
• manual browser validation in editor mode and render-only mode
• test -f "/workspace/llm-agent/skills/graph-wizard/SKILL.md" && printf 'skill file present\n'
• which agent && agent --version && agent --help
• agent --print --trust --workspace /workspace "Use the graph-wizard skill to generate a graph of a network topology for how AWS Lambda functions work and show the logging infrastructure. Put the output in javascript/max-graph-editor/out/cli-skill-test." (blocked by authentication requirement)

Walkthrough

graph_wizard_lambda_topology_demo.mp4
Graph wizard editor mode
Graph wizard render mode
Generated Lambda topology PNG

Notes

• graph-wizard uses OpenAI when OPENAI_API_KEY is configured and otherwise falls back to a deterministic heuristic drafter
• the tested output bundle includes maxGraph XML, mxGraph-compatible XML, a rendered PNG, a diagram report, and a markdown summary
• the new skill teaches agents when to use the workflow, which files matter, the Bun commands to run, and how to validate output quality
• Cursor Agent installation succeeded, but a real CLI-driven skill run requires interactive agent login or CURSOR_API_KEY

_{To show artifacts inline, enable in settings.}