Design Farmer logo

Design Farmer

Community
hakilee
design-farmer

Automated design system construction from repository analysis to production-ready implementation. Analyzes codebases, extracts design patterns, builds token hierarchies with OKLCH color management, implements accessible components with tests, and verifies through multi-reviewer panels. Use when: "build design system", "design tokens", "component library", "design-farmer"

Overview

Publisherhakilee
Repositorydesign-farmer
Skill namedesign-farmer
Stars
61
Forks
13
Bundled files
26
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 26 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by hakilee on GitHub. Read the source before you install it.

Installation

Install the Design Farmer AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/hakilee/design-farmer.git /tmp/design-farmer
mkdir -p .claude/skills
cp -r /tmp/design-farmer/skills/design-farmer .claude/skills/design-farmer
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Design Farmer in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Design Farmer on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Design Farmer is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Design Farmer

From seed to system — cultivate a production-ready design system from any codebase.

Design Farmer analyzes your repository, extracts existing design patterns, and grows them into a structured, accessible, OKLCH-native design system with tokens, components, tests, and documentation.


Version Check

Before doing anything else, run the version check silently:

bash
_DF_UPDATE=$(bash ~/.claude/skills/design-farmer/bin/version-check 2>/dev/null || true)

If _DF_UPDATE starts with UPGRADE_AVAILABLE:

  1. Parse the output: UPGRADE_AVAILABLE <current> <latest>
  2. Ask the user via AskUserQuestion:
Design Farmer **v{latest}** is available (you're on v{current}).

Options:
- A) Update now — run the installer, then restart
- B) Continue with current version
- C) Skip — don't ask again this session
  • If A: output the install command below and stop. The user must re-invoke after updating.
bash
curl -fsSL https://raw.githubusercontent.com/ohprettyhak/design-farmer/main/install.sh | bash
  • If B or C: continue to the next section.

If the script is missing, fails, or times out, continue silently — never block on a version check failure.


Bundle Integrity Gate

This skill is distributed as a bundle: the router SKILL.md, phase files under phases/, and companion docs under docs/ must all be present.

Before starting work, Read phases/operational-notes.md to verify the bundle is accessible.

Before each phase, Read the corresponding phase file using the Read tool with its full path (e.g., Read("~/.claude/skills/design-farmer/phases/phase-0-preflight.md")). If the Read call fails (file not found), the bundle is incomplete — STOP immediately and report:

Status: BLOCKED
Reason: Incomplete Design Farmer bundle — required file missing: {path}
AskUserQuestion: "The installed Design Farmer skill bundle is incomplete (`{path}` is missing). Please reinstall or provide the missing file before I continue."

Important: Do NOT use Glob or Search tools with patterns like phases/*.md to verify phase files — these patterns may return zero results due to tool-specific path resolution behavior. Always use Read with the exact file path.

Do NOT guess missing phase behavior from memory.


Phase Architecture

This skill is decomposed into specialized phase files under phases/. Load only the file for the current phase plus any explicitly referenced companion file.

PhaseFilePurpose
0phases/phase-0-preflight.mdDetect project topology, frameworks, package manager, and existing design artifacts
1phases/phase-1-discovery.mdRun the one-question-at-a-time discovery interview and build DesignFarmerConfig
2phases/phase-2-repo-analysis.mdAssess design maturity, inventory components, and extract repository patterns
3phases/phase-3-pattern-extraction.mdNormalize color/typography/spacing patterns into token-ready design primitives
3.5phases/phase-3.5-visual-preview.mdGenerate and review a design preview before implementation begins
4phases/phase-4-architecture.mdDefine token hierarchy, directory structure, build pipeline, and CSS layering
4bphases/phase-4b-theming.mdDefine theme system, provider implementation, dark mode, and styling approach
4.5phases/phase-4.5-design-source-of-truth.mdGenerate and maintain DESIGN.md as the persistent design reference
5phases/phase-5-tokens.mdImplement primitive/semantic/component tokens and token tests
6phases/phase-6-components.mdImplement components in dependency order
7phases/phase-7-storybook.mdSet up Storybook and visual documentation when the user wants it
8phases/phase-8-review.mdRun multi-reviewer verification and aggregate risk
8.5phases/phase-8.5-design-review.mdRun rendered visual QA with browser tooling or manual fallback
9phases/phase-9-documentation.mdGenerate docs, final verification output, and completion report
10phases/phase-10-integration.mdIntegrate the design system into the application when requested
11phases/phase-11-readiness-handoff.mdProduce release-readiness verification and handoff output

Reference Files

FilePurpose
phases/operational-notes.mdDelegation strategy, escalation rules, OKLCH reference, forbidden patterns
docs/PHASE-INDEX.mdConcise phase responsibilities and handoff map
docs/QUALITY-GATES.mdStructural and behavioral quality gates for maintainers
docs/MAINTENANCE.mdUpdate workflow, anti-drift checks, and contribution checklist
docs/EXAMPLES-GALLERY.mdExample outcomes and reference scenarios
examples/DESIGN.mdFully filled-in DESIGN.md example (Nova UI) for greenfield reference

AskUserQuestion Tool Usage — MANDATORY

CRITICAL: When a phase file says "Via AskUserQuestion, ask:" or "ask via AskUserQuestion:", you MUST call the AskUserQuestion tool. Do NOT output the question as regular text or markdown. The blockquoted text following the instruction is the CONTENT for the tool call, not text to display.

Every AskUserQuestion call must:

  1. Use the AskUserQuestion tool with structured questions, header, and options parameters.
  2. Wait for the tool to return the user's response before taking any further action.
  3. Never proceed past a → STOP marker until the user's response has been received from the tool.

If you output question text without calling the tool, the user cannot respond and the pipeline breaks.


Voice & Tone

  • Speak like a senior design engineer pairing with the user — direct, specific, opinionated with reasoning.
  • Name the file, the token, the component. Never speak in abstractions when a concrete path exists.
  • One decision per question. Never batch multiple choices into a single prompt.
  • Recommend the complete option. With AI assistance, thoroughness costs seconds more than shortcuts.
  • When uncertain about the user's intent, ask. When certain, act and explain.

Completion Status Protocol

Every phase concludes with one of:

  • DONE — Phase complete with evidence (files created, tests passing, screenshots captured).
  • DONE_WITH_CONCERNS — Phase complete but issues flagged for user awareness.
  • BLOCKED — Cannot proceed. State what is missing and what would unblock.
  • NEEDS_CONTEXT — Missing information that only the user can provide. Ask one question.

Fallback & Degradation Protocol

Every phase that delegates to an Agent or uses an external tool MUST define a fallback path. The pipeline NEVER silently fails.

Compatibility note: some runtimes represent delegation explicitly as Agent(prompt="...").

Degradation Pattern

Try primary path:
  If success → continue to next phase
  If failure (timeout, token limit, tool unavailable):
    Log: "[DEGRADATION] Phase {N}: {primary} failed ({reason}). Using {fallback}."
    Execute fallback path
    If fallback succeeds → continue with DONE_WITH_CONCERNS
    If fallback also fails:
      Escalate to user with BLOCKED status
      AskUserQuestion: "Both primary and fallback failed for {phase}. Error: {error}. How to proceed?"

Fallback Registry

PhasePrimary PathFallback Path
Phase 2 (Analysis)Structured repository analysis pass (specialized delegation preferred)Direct Grep/Glob scanning with reduced depth
Phase 3 (Extraction)Specialized analysis passManual OKLCH extraction via inline math
Phase 3.5 (Visual Preview)Generated preview HTMLText-only preview summary and user approval gate
Phase 4.5 (DESIGN.md)Structured design-document draftingWrite DESIGN.md directly with the documented template
Phase 5 (Tokens)Specialized implementation passImplement tokens directly with Edit/Write
Phase 6 (Components)Specialized implementation passImplement components directly with Edit/Write, one at a time
Phase 7 (Storybook)storybook init via the detected package managerManual .storybook config + story file generation
Phase 8 (Review)5 specialized reviewer passes (parallel when supported)Sequential review with combined criteria
Phase 8.5 (Visual QA)Headless browser screenshotsManual verification prompt with user-provided screenshots
Phase 9 (Docs)Structured documentation draftingWrite docs directly with Write tool
Phase 10 (Integration)Structured application changesGuided step-by-step instructions for manual execution
Phase 11 (Readiness)Automated verification + readiness handoffManual readiness report with explicit failed-gate summary

Execution Flow

Execute phases sequentially. Before starting each phase:

  1. Read phases/operational-notes.md once per run if you have not already loaded it.
  2. Read the corresponding phase file.
  3. Follow the instructions in that file exactly.
  4. Conclude the phase with a completion status.
  5. Proceed only when that phase’s gates allow it.

Phase Sequence

Phase 0: Pre-flight
    |
Phase 1: Discovery Interview
    |
Phase 2: Repository Analysis
    |
Phase 3: Design Pattern Extraction & OKLCH Conversion
    |
Phase 3.5: Visual Preview
    |
Phase 4: Architecture Design
    |
Phase 4b: Theme & Styling
    |
Phase 4.5: Design Source of Truth (DESIGN.md)
    |
Phase 5: Token Implementation
    |
Phase 6: Component Implementation
    |
Phase 7: Storybook Integration (optional — user chooses)
    |
Phase 8: Multi-Reviewer Verification
    |
Phase 8.5: Design Review (Live Visual QA) (if browser tooling or screenshots are available)
    |
Phase 9: Documentation & Completion
    |
Phase 10: App Integration (optional — user chooses)
    |
Phase 11: Release Readiness & Handoff

Cross-Phase Contracts

  • DesignFarmerConfig (from Phase 1) is passed to all subsequent phases. Persisted to {systemPath}/.design-farmer/config.json for context-window resilience.
  • Pipeline state tracking: Completed and degraded phases append their IDs to completedPhases in config.json. User-optional skipped phases record dedicated skip state instead and do not append. createdAt is set once in Phase 1. Phase 8 writes lastReviewScore (0–10) and lastReviewDate. Phase 0 displays this state during re-entry.
  • Design Maturity (from Phase 2) is written to DesignFarmerConfig.designMaturity and determines the implementation path in Phases 3, 3.5, 5, and 6. Phase 0 re-entry may set a preliminary designMaturity from user input; Phase 2's formal assessment overrides this value.
  • Existing DESIGN.md detected in Phase 0 triggers a re-entry prompt — importing it as context continues to Phase 1 with pre-filled defaults. Critical discovery gates (scope/headless/theme decisions) must still be explicitly confirmed.
  • DESIGN.md source classification: Phase 0 distinguishes internal-canonical vs external-context. Readable third-party docs are context (not corruption); only unreadable files are treated as corrupted.
  • Early DESIGN.md draft (from Phase 3) captures extraction results to bridge the Phase 3→4.5 context gap. Phase 4.5 merges this draft into the final version.
  • Phase 3.5 is a hard gate before Phase 4. Preview HTML ({systemPath}/.design-farmer/design-preview.html) is mandatory for GREENFIELD, opt-in for EMERGING/MATURE. When skipped, the opt-in gate (3.5.0) handles text-only approval directly — the error-state Fallback Path (3.5.3) is reserved for generation failures only. The generatePreview field records the choice.
  • Phase 4b is a continuation of Phase 4 — load it immediately after Phase 4 completes, before Phase 4.5.
  • DESIGN.md (from Phase 4.5) is the persistent design source of truth for Phases 5–11.
  • Fix Loop Protocol (from operational-notes.md) is mandatory at implementation checkpoints in Phases 5, 6, 7, 9, 10, and 11. Each phase must pass typecheck/lint/build/test before proceeding. The loop retries up to 5 times, then escalates to BLOCKED.
  • Semantic-token-only rule: Components must NEVER consume primitive tokens directly.
  • Completion statuses are mandatory: Every phase must end with DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT.
  • User-question gating: Discovery interview questions are one-at-a-time. Other AskUserQuestion calls require user response before proceeding.
  • Final completion requires explicit verification evidence from Phase 9.2 and a readiness handoff report from Phase 11.
  • Optional phase skip policy: Two skip mechanisms exist:
    • Context-first re-entry (Phase 0→1): Existing DESIGN.md is imported as context and does not bypass Phases 1–4.5 by default. Legacy skippedPhases entries remain backward-compatible when present from older runs.
    • User-optional phase skip: Phases 7 (Storybook), 8.5 (Design Review), and 10 (App Integration) may be skipped by user choice or environment constraints. Skip status is tracked via dedicated flags: storybookSkipped: true (Phase 7), visualQASkipped: true (Phase 8.5), integrationStatus: "skipped" (Phase 10). Skipped user-optional phases do NOT append to completedPhases. Degraded execution (e.g., manual fallback) DOES append. Phase 8.5 also records visualQAMode: 'auto' | 'manual' | 'skipped' to distinguish execution quality.

Maintenance Note

This router is the canonical runtime entrypoint. If phase boundaries, file names, or quality criteria change, update the corresponding phase file, docs/PHASE-INDEX.md, docs/QUALITY-GATES.md, docs/MAINTENANCE.md, and scripts/validate-skill-md.sh in the same PR.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Design Farmer AI skill do?

Automated design system construction from repository analysis to production-ready implementation. Analyzes codebases, extracts design patterns, builds token hierarchies with OKLCH color management, implements accessible components with tests, and verifies through multi-reviewer panels. Use when: "build design system", "design tokens", "component library", "design-farmer"

Why use Design Farmer on TypingMind?

Because you install it once and use it with any model. Design Farmer is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Design Farmer in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/hakilee/design-farmer/tree/main/skills/design-farmer. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Design Farmer?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Design Farmer?

As many as you like. As long as a model supports skills, you can use Design Farmer with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Design Farmer AI skill free?

It is published on GitHub by hakilee. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇