Logo-design skill for Claude ships with a 1,400-file SVG reference library and Python tools that audit an SVG against logo principles

A founder who needs a logo has two starting points today: hire a designer for a discovery brief that runs for a week, or open a text-to-image tool and hope the result is still legible when scaled down to 16 by 16 pixels for a browser tab. A new open-source Claude skill by an author who goes by kaankiziltug adds a third: a disciplined logo-design workflow the model runs against a 1,400-file reference library, with automatic checks for the tests a designer would run by hand.

The skill lives at kaankiziltug/logo-design-skill on GitHub, published under the MIT licence for the skill text, scripts, templates and catalog data. The repository was created on 2026-09-26 and has 97 stars in its first day. The trademarked logos in the reference library sit under their own separate notice and are marked for reference and education only.

The library and the twelve principles

The skill loads on demand from a set of reference files. The principles reference covers 12 logo-design principles, discovery briefs, mark types, visual techniques such as geometry, grids, optical corrections and negative space, colour, typography, SVG construction, a testing checklist, presentation and delivery, identity systems, redesigns and critique.

The reference library ships with more than 1,400 real-world SVG logos from about 1,200 brands, classified by mark type, technique, geometry, subject, typography, mood and industry. The author says the library exists so the model can study construction, map category conventions and avoid look-alikes, and states plainly that it is never for copying. According to the README, 24 percent of the library are abstract marks, 21 percent combination, 20 percent pictorial, 15 percent letterform and 10 percent wordmark, and 141 files are flagged as exemplary teaching examples.

The checkpoint, and the tools that back it

The skill follows a fixed sequence and stops at a checkpoint. It runs discovery, researches the category in the library, produces 8 to 12 concept one-liners, builds 3 as SVGs, tests and refines them, and then shows the three concepts as one greyscale overview image at 64, 32 and 16 pixels with a written recommendation. Nothing else is produced until the user picks a direction, on the grounds that the kit is most of the work and only makes sense once a direction is chosen.

The Python scripts that back the checkpoint are dependency-free and run from the standard library. svg_audit.py checks structure, colour count, complexity, near-miss angles, tiny details and centring. preview_sheet.py produces an HTML test sheet with size, background, treatment and shelf tests. presentation_board.py builds a client presentation with six industry-specific mockups per concept. search_library.py queries the library by technique or industry. export_variants.py writes black, white, mono, square, favicon, app-icon and web-icon versions in one call.

What a working reader can do with it

The skill runs from the Claude Code plugin marketplace, or as a manual copy into a personal or per-project skills folder, or as an upload to Claude Desktop under Settings, Capabilities, Skills. The README lists 18 fictional briefs worked end-to-end, from a specialty coffee roaster and a fintech to a physiotherapy clinic and a telehealth app, each with its concept overview, chosen mark and reasons.

Traction is thin at 97 stars in one day, so the reason to look is the skill itself. A founder or a marketer using Claude can now go from a written brief to a presented concept with size and contrast checks already run, and the checkpoint stops the model before it spends hours producing a kit for a direction the user has not approved.

Source

• Repository and README: kaankiziltug/logo-design-skill on GitHub