A public Claude Code plugin marketplace by Reza Nirumand, hosting agentic skills and agents as installable plugins.
Add the marketplace once:
/plugin marketplace add nirumand/skills
Then install any plugin from it:
/plugin install diagram-tools@nirumand-skills
To update later:
/plugin marketplace update nirumand-skills
Create, review, and report on software architecture diagrams following the 12 rules from INNOQ's "Better Architecture Diagrams" (Gernot Starke).
| Component | Type | What it does |
|---|---|---|
diagram-create |
Skill | Creates architecture diagrams (Mermaid/PlantUML) from requirements, with a mandatory self-verification checklist before delivery. Triggers on "create/draw/generate a diagram". |
diagram-review |
Skill | Reviews an existing diagram (image, Mermaid, PlantUML, draw.io) against all 12 rules — findings table, prioritized fixes, and a rule-compliant redraw. Triggers on "review/check/critique this diagram". |
diagram-report |
Skill | Renders a completed review as a standalone shareable HTML report: original, findings, fixes, and redraw side by side. |
diagram-reviewer |
Agent | Reviews one diagram and returns a compact findings report. Designed for bulk reviews — fan out one agent per diagram in parallel so image reads stay out of the main context. |
Usage examples:
> Create a context diagram for our payment service
> Review docs/architecture/overview.png — is this diagram good?
> Review all diagrams in docs/ and rank them
> Generate an HTML report for that review
How the pieces fit together:
diagram-createholds the shared rulebook (references/diagram-rules.md) — the 12 rules, checklist, and review weighting used by everything else.diagram-reviewreviews a single diagram inline; for many diagrams it fans outdiagram-revieweragents in parallel.diagram-reportturns review output into a portable HTML file (Mermaid redraw rendered live via CDN).
Attribution:
This plugin is drafted from the article "Better Architecture Diagrams — 12 rules" by Gernot Starke (INNOQ). The 12 rules, the checklist, and the review weighting embedded in the skills are a condensed adaptation of that work — full credit for the underlying methodology belongs to its author. If you find the plugin useful, read the original article.
.
├── .claude-plugin/
│ └── marketplace.json # Marketplace manifest (lists all plugins)
├── plugins/
│ └── diagram-tools/
│ ├── .claude-plugin/
│ │ └── plugin.json # Plugin manifest
│ ├── agents/
│ │ └── diagram-reviewer.md
│ └── skills/
│ ├── diagram-create/
│ │ ├── SKILL.md
│ │ └── references/diagram-rules.md
│ ├── diagram-review/
│ │ └── SKILL.md
│ └── diagram-report/
│ ├── SKILL.md
│ └── assets/report-template.html
├── evals/ # Eval suite (dev tooling, not shipped with the plugin)
│ ├── fixtures/ # Test diagrams and pre-baked review
│ └── <skill-name>/ # evals.json + trigger-evals.json per skill
├── LICENSE
└── README.md
Each skill has a thin eval suite under evals/: trigger-accuracy prompts (should/should-not invoke the skill, with deliberate near-misses) and end-to-end scenarios with objective format assertions (12-row findings table, legend present, no unfilled template placeholders). Subjective quality is left to human review by design. See evals/README.md for how to run them.
- Create
plugins/<plugin-name>/with a.claude-plugin/plugin.jsonmanifest. - Add
skills/,agents/,commands/, orhooks/directories as needed — Claude Code auto-discovers them. - Register the plugin in
.claude-plugin/marketplace.jsonunderplugins. - Reference files across skills with
${CLAUDE_PLUGIN_ROOT}/...— never absolute paths, so the plugin works wherever it is installed.
MIT © Reza Nirumand