Is there a usable collection of Claude skills for generating diagrams and briefs? Specifically: is there an ERD skill?
2026-07-31 · complete · confidence: high
# Claude skills for diagrams and briefs — what actually exists
Question: is there a usable collection of Claude skills for generating diagrams and briefs? And specifically — is there an ERD skill?
Short answer: the ecosystem is real but far thinner than the aggregator sites suggest, and there is no ERD skill. For anyone modelling data, that gap is the finding.
A skill is a folder with a SKILL.md file: YAML frontmatter plus instructions. Claude loads it through progressive disclosure — metadata first (~100 tokens), full instructions only when it judges the skill relevant, bundled resources and code only when needed.
Install by cloning into ~/.claude/skills/ or .claude/skills/, or via the plugin marketplace:
/plugin marketplace add anthropics/skills /plugin add /path/to/skill-directory
Skills are not MCP servers. A skill activates on context and injects instructions; a server exposes callable tools. Both appear on the same "diagram tooling" lists, which is a common source of confusion when picking one.
| Repository | Stars | Licence | Last push | What it is |
|---|---|---|---|---|
anthropics/skills | 165,392 | Apache-2.0 (+ source-available) | 2026-07-24 | The official repo |
obra/superpowers | 264,253 | MIT | 2026-07-28 | Largest community collection |
travisvn/awesome-claude-skills | 14,438 | — | 2026-04-28 | Curated list |
karanb192/awesome-claude-skills | 446 | MIT | 2026-07-28 | Curated list, 50+ verified |
jgraph/drawio-mcp | 5,046 | Apache-2.0 | — | draw.io MCP server (not a skill) |
The official repo, in full — 17 skills: algorithmic-art · brand-guidelines · canvas-design · claude-api · doc-coauthoring · docx · frontend-design · internal-comms · mcp-builder · pdf · pptx · skill-creator · slack-gif-creator · theme-factory · web-artifacts-builder · webapp-testing · xlsx.
Licensing splits: most are Apache-2.0; the four document skills (docx, pdf, pptx, xlsx) are source-available, not open source — shared as reference implementations of complex production skills.
⚠️ Not one of the 17 generates a diagram. canvas-design and algorithmic-art are visual, but neither turns a model into an ERD, a lineage graph or a flowchart.
Several sites confidently name specific diagram skills. Checked against GitHub:
| Cited as | Reality |
|---|---|
ryanquinn3/excalidraw-skill | 404 — does not exist |
jgtolentino/drawio-diagrams-enhanced | 404 — does not exist |
jgraph/drawio-mcp | ✅ exists, 5,046★ — but an MCP server, not a skill |
A GitHub search for diagram skills returns essentially two results: tEhThing/claude-skills-diagramming (2 stars) and ergonzal22/claude-skill-diagram-generator (0 stars).
That is the honest state of it. The blog posts and marketplace pages describing a mature three-skill landscape are describing something that mostly is not there — or is there under a different mechanism (MCP servers, built-in connectors) than the word "skill" implies.
Method note: every claim here was checked against the GitHub API rather than taken from the write-ups. Two of three named repos evaporated on contact. Worth remembering for any tooling research — the aggregators are not a source, they are a lead.
Searches for ERD, entity-relationship and data-model skills return nothing usable.
Why that gap is interesting rather than disappointing. The existing visual skills all start from a description — you tell Claude what to draw, it emits a picture. That is Model as Drawing with a language model attached: the output looks like a diagram and generates nothing, validates against nothing, and drifts the moment reality changes.
An ERD skill worth having would start from the model, not from a prompt:
None of that is exotic. It is exactly what the generators behind this site already do — the missing piece is the thin SKILL.md wrapper that makes them available inside a conversation.
⚠️ The distinction that matters. A skill that draws an ERD from a prompt is a nicer way to make a picture. A skill that generates one from metadata is a view of a model. They look identical in a demo and behave differently in month three.
Nothing purpose-built. The nearest neighbours are doc-coauthoring (official) and the document skills for output formats.
The gap is the same shape as the ERD one: a brief in our sense is markdown with a metadata header — a small structured artifact that is reusable, queryable and citable, not a prompt typed once into a chat window. That is a convention plus a template, not a hard skill to write.
1. The ecosystem is real but thin. Two enormous repos (anthropics/skills, obra/superpowers) and a handful of curated lists — but essentially nothing for diagrams.
2. Not one of the 17 official skills generates a diagram.
3. Two of the three widely-cited diagram skills do not exist. Verify before citing.
4. jgraph/drawio-mcp is real and useful — but a server, not a skill. Different install, different activation.
5. There is no ERD skill. For a data-modelling audience, that is an unoccupied space.
6. The existing skills draw from prompts, not models. Which is precisely the anti-pattern this site argues against.
7. We already have the generators. Turning them into skills is packaging work, not invention.
obra/superpowers (264k★) for a diagram skill hiding under another name