Diagramkit

diagramkit — create, render, embed, and audit diagrams

Owns the diagram lifecycle in any repo that has diagramkit installed: author a new diagram, re-render a source, embed a rendered diagram into content, or run a repo-wide audit that validates every diagram and fixes contrast/WCAG issues. Writes to the local repo and runs the local diagramkit CLI — it never commits, pushes, or publishes.

This skill is a thin router. It does not restate diagramkit's version-sensitive knowledge — engine selection, per-engine palettes, readability budgets, config schema, and the exact <picture> markup all live in the consumer's installed package and drift the moment a version ships. The real authoring is always delegated: read and follow node_modules/diagramkit/skills/diagramkit-<engine>/SKILL.md. Broader Pagesmith content/site/deploy work routes to /adk:pagesmith.

The full operating contract lives in this skill folder — read these as you need them:

Aspect	File
How you work (voice, delegation discipline, output shape)	persona.md
The phased process + Workflow orchestration	workflow.md
Hard rules + refusals + safety	rules.md
Route dispatch (new-diagram / render / embed / audit)	dispatch.md

Quick start

1. Verify + materialize (Phase 0 in workflow.md). Confirm diagramkit is installed and Chromium is warm with npx diagramkit doctor. Materialize the per-engine skill stubs with npx diagramkit skills install — if that subcommand isn't in the installed version, fall back to following node_modules/diagramkit/skills/diagramkit-setup/SKILL.md.
2. Read dispatch.md and classify the input into one route: new-diagram, render, embed, or audit.
3. Delegate the authoring. For a new diagram, select the engine via node_modules/diagramkit/skills/diagramkit-auto/SKILL.md, then author by following that engine's node_modules/diagramkit/skills/diagramkit-<engine>/SKILL.md. Never invent a palette or budget here — the installed skill is the source of truth.
4. Render + validate. npx diagramkit render the source, then npx diagramkit validate (add --fail-on <severity> / --scope-dir <dir> if the installed version supports them; else fall back to a plain validate). A failing gate stops you (rules.md).
5. Embed / report. For the embed route, produce the light/dark <picture> markup per the installed skill's guidance and place it in the target content file. Report the sources, renders, validation results, and any embed — nothing is committed or published.

Workflow is the default for the repo-wide audit

"Always have a workflow." The repo-wide audit gets the Workflow in workflow.md: fan out one agent per engine (or per content area), each running diagramkit validate over its slice, applying contrast/WCAG fixes and re-rendering stale SVGs by following its engine's installed skill, then a final repo-wide diagramkit validate gate confirms the fixes held. A single new diagram or a one-source re-render is done inline by delegating to the one relevant engine skill — say you skipped the Workflow.

Modes

• default — infer the route from the input (dispatch.md): author a new diagram, re-render a source, embed into content, or audit the repo. Verify → delegate → render → validate → report. Writes local files only.
• --engine <name> — skip auto-selection; author with the named engine by following node_modules/diagramkit/skills/diagramkit-<engine>/SKILL.md.
• --audit — force the repo-wide audit: fan out per engine/content area via the Workflow tool, fix contrast/WCAG, re-render stale SVGs, re-validate. Follows diagramkit's own review flow (node_modules/diagramkit/skills/diagramkit-review/SKILL.md).
• --embed <content-path> — produce the light/dark <picture> embed for the target content file per the installed skill; route broader page authoring to /adk:pagesmith.
• --scope-dir <dir> / --fail-on <severity> — passed through to diagramkit validate when the installed version supports them; otherwise fall back to a plain validate scoped to that subtree, and say so.
• -i — confirm the engine choice, the audit scope, and each contrast/WCAG fix before applying it.
• --deep — use a stronger reasoning profile; auto-select for a large audit or an ambiguous engine choice.