Your coding agent's diagrams, finally worth looking at

What it does

SeeCode is an agent skill: you ask your coding assistant for a diagram in plain language (“show how checkout works”), and it emits a compact JSON spec that SeeCode renders as a single self-contained HTML file. The result isn’t a static PNG — it’s a diagram you can hover, click, and walk through, with animations that trace flows, play sequences message by message, or grow charts. When you’re done admiring it, export to SVG, PNG, GIF, or MP4.

The interesting bit

The boring part is the value: charts pull their numbers from a CSV or JSON file, so data never passes through the model — no hallucinated axis values. And the “architecture delta” type, which visualizes what a change adds, removes, or alters, is a genuinely novel fit for code-review-adjacent work.

Key highlights

• 45 diagram types, from architecture, sequence, and ER models to Wardley maps, fishbones, and isometric exploded views of, apparently, espresso machines
• Every type animates its own way but can also be fully static
• Interactive viewer features: hover traces, focus panels, route journeys
• Zero dependencies, MIT licensed
• Works as a Claude Code plugin or a generic Agent Skill for Cursor, Codex, OpenCode, and other hosts
• Themes including dark mode and custom branding

Caveats

• The breadth (45 types) suggests depth varies; the README doesn’t say which types are more polished than others
• It’s glue between your agent and a renderer — the output quality depends on how well your agent writes the spec

Verdict

If you live in Claude Code or another agent host and regularly explain systems to humans, this is a strong upgrade over pasting Mermaid into a README. If you already have a diagrams-as-code pipeline you’re happy with, it’s a nice-to-have with a very fun demo reel.