Design#Branding

logo-design-skill: the designer workflow as an Agent Skill, with a 1,400-logo reference library

Open skill for the logo workflow: vector sketches from 8–12 directions, 16 px, one-colour and shelf tests, backed by a 1,400-logo reference library.

Skill details

Install
npx skills add kaankiziltug/logo-design-skill --skill logo-design

The familiar failure of AI-drawn logos: the hero image looks fine, the browser-favicon size turns to mush, and the one-colour version is unrecognizable. logo-design-skill, featured by GitHub_Daily on November 13, 2026, targets exactly that gap. It writes the professional designer’s workflow as an Agent Skill — from the first brief through directions, sketches, testing and a full delivery package. GitHub’s API reported 6,887 stars on November 14, 2026.

Core features

  • A structured process: generate 8 to 12 directions, draw 3 as vector sketches, stop for the user’s direction pick, then produce colour, lockups and a full brand guide — the sketch confirmation stays in human hands.
  • Scaling and degradation tests: a 16 px pixel test, one-colour, reversed, squint, mirror, in-context placement, and a competitor shelf test that puts the new mark next to similar brands to catch collisions.
  • A 1,400+ real-logo reference library: SVG marks classified by type, technique and industry; consult comparable designs before ideating, then reuse the shelf to check whether a new mark resembles an existing brand.
  • A complete delivery: final SVG, presentation sheets, and the icon sizes a website and app actually need.
  • Broad agent support: built on the open Agent Skills format (a folder with a SKILL.md); the README lists Claude Code (plugin marketplace or manual), Claude.ai / Claude Desktop, Gemini CLI, Codex CLI, Cursor and GitHub Copilot.

Typical use cases

  • Indie developers and small teams: a complete identity that survives scaling tests, not a one-off hero image, without a designer budget.
  • Pre-redesign collision checks: run the shelf test against the reference library before committing to a new direction, cutting rework on lookalike marks.
  • Quality control on bulk generation: whichever model produces the drafts, the test checklist works as an acceptance standard to run them through. Its real value is moving veto criteria such as “survives 16 px” to the sketch stage instead of after delivery.

Quick start

For Claude Code, the README recommends the plugin marketplace; the manual route copies the skill folder into the project:

cp -r logo-design-skill/skills/logo-design .claude/skills/logo-design

Other agents use their own directories: Gemini CLI at ~/.gemini/skills/logo-design (or ~/.agents/skills/), Codex CLI at ~/.codex/skills/logo-design. The .claude-plugin/ folder is only used by Claude Code and ignored elsewhere; confirm your agent supports the Agent Skills format first. If your workflow already keeps design tokens in the engineering path, compare design-extract: it pulls a design system out of an existing site, complementing this skill’s generate-side workflow.

Summary

The value here is not “can draw a logo” — it is the designer’s veto list written as test steps: scale down, drop to one colour, reverse it, put it on the shelf next to competitors; each step can kill a direction before launch-day rework. GitHub’s API reported MIT, HTML, 6,887 stars, an unarchived repository, last pushed November 13, 2026, on November 14, 2026.

This entry is based on the README and skill docs; we did not run the full design flow. The reference library’s classification quality and real-world pass rates are worth testing on one of your own projects.