nadicode-ai-designer

name	nadicode-ai-designer
description	Generates visual assets (icons, illustrations, logos, hero images, background patterns, section dividers, Lottie animations, short video clips) by routing each request to the best AI provider and running a provider-tuned prompt. Use when the user asks for any graphic, image, icon, SVG, logo, hero, background, illustration, pattern, avatar, animation, or video asset. Routes between agent-svg (the calling agent authors SVG directly, with Recraft --format svg as the deterministic fallback for non-LLM agents), Recraft (vector and Lottie), FLUX (artistic raster), OpenAI Images (photorealistic), Ideogram (typography-heavy), Nano Banana (general raster), and Veo (video). Includes interactive credential setup via scripts/setup.js when providers are unconfigured.
license	MIT
compatibility	Requires Node.js >= 20 on PATH. Tested install paths for Claude Code, Codex, Cursor, GitHub Copilot CLI, Gemini CLI, Goose, OpenHands, Amp, Windsurf, JetBrains Junie, Continue, OpenCode, and Aider (read-only).
allowed-tools	Bash(node scripts/doctor.js:*), Bash(node scripts/setup.js:*), Bash(node scripts/list-jobs.js:*), Bash(node scripts/generate-flux.js:*), Bash(node scripts/generate-recraft.js:*), Bash(node scripts/generate-openai.js:*), Bash(node scripts/generate-ideogram.js:*), Bash(node scripts/generate-nanobanana.js:*), Bash(node scripts/generate-veo.js:*)

nadicode-ai-designer

Generate visual assets via the right AI provider for each request.

Invocation

This skill follows the Agent Skills open standard. It is auto-invoked when the user asks for any graphic, image, icon, SVG, logo, hero, background, illustration, pattern, avatar, animation, or video asset. Agents that support slash commands also expose it as /nadicode-ai-designer. No explicit invocation is required: describing the asset in natural language is enough.

Representative user phrases that trigger this skill: "make me a search icon", "generate a hero background", "I need a logo for X", "draw a section divider", "create an avatar", "give me a Lottie loading animation", "render a short video clip of Y". In every case, the calling agent reads this file, picks a provider via the decision flow below, and either authors SVG directly (agent-svg path) or invokes one of the bundled scripts/generate-*.js binaries.

First-time setup

Before the first generation, check which providers are configured:

node scripts/doctor.js --json

If the provider for the user's request is unconfigured, offer interactive setup in chat:

1. Tell the user which token they need and where to get it:
   • FLUX / Recraft → REPLICATE_API_TOKEN from https://replicate.com/account/api-tokens
   • OpenAI Images → OPENAI_API_KEY from https://platform.openai.com/api-keys
   • Ideogram → IDEOGRAM_API_KEY from https://ideogram.ai/account/api
   • Nano Banana / Veo → GOOGLE_API_KEY from https://aistudio.google.com/app/apikey
2. Ask the user to paste the token directly into the chat.
3. Run:

   node scripts/setup.js --provider <name> --token <pasted-value> --json
   

   The script validates the token against the provider's API and persists it to .nadicode-ai-designer/config.json (mode 0600, gitignored).
4. Re-run doctor.js to confirm.

Do not ask the user to set env vars and restart the session. Drive setup inline.

The agent-svg path needs no token: if the calling agent can author SVG code, it does so directly. Agents that cannot author SVG route to Recraft's --format svg mode as the deterministic fallback (shared REPLICATE_API_TOKEN).

Decision flow

Read references/provider-routing.md for the full decision tree. Summary:

Want...	Provider	How
Icon, geometric pattern, section divider (SVG code)	agent-svg	Read references/agent-svg.md, then write the SVG directly to the target file. If your agent cannot author SVG, fall back to node scripts/generate-recraft.js --format svg — but note the fallback produces illustrated line-art SVG (larger, fixed colors), not the minified 24x24 currentColor icon the LLM-author path produces. See agent-svg.md "Quality tradeoff" for details.
Vector illustration, Lottie animation	Recraft	node scripts/generate-recraft.js
Artistic hero background, abstract raster	FLUX	node scripts/generate-flux.js
Photorealistic photo, avatar, product shot	OpenAI Images	node scripts/generate-openai.js
Logo, wordmark, typography-heavy graphic	Ideogram	node scripts/generate-ideogram.js
General-purpose raster image	Nano Banana	node scripts/generate-nanobanana.js
Short video clip (MP4)	Veo	node scripts/generate-veo.js

When NOT to use this skill

• Editing existing images — this skill only generates. Use a different tool for crop, resize, filters, or inpainting.
• UI component code — this skill generates visual assets (files), not React/Vue components.
• 3D models, CAD, datasets, charts from data — out of scope.
• Photo restoration, face swap, deepfakes — out of scope.

Prompting per provider

Before calling any script, load the matching reference file for best-practice prompting:

Provider	Reference
agent-svg (agent-written, or Recraft --format svg fallback)	references/agent-svg.md
Recraft (vector / Lottie)	references/recraft-prompting.md
FLUX (no negative prompts; use hex colors tied to objects)	references/flux-prompting.md
OpenAI Images	references/openai-prompting.md
Ideogram (exact text in quotes, first 150 words)	references/ideogram-prompting.md
Nano Banana	references/nanobanana-prompting.md
Veo (video)	references/veo-prompting.md
Lottie (via Recraft JSON)	references/lottie-animation.md
All-provider capabilities + pricing overview	references/provider-knowledge.md

Running a generator script

All generator scripts accept --json for machine-readable output:

node scripts/generate-flux.js \
  --prompt "cinematic deep blue fintech hero, soft atmospheric glow, hex #0a0a0f base" \
  --size 1920x1080 \
  --output ./public/hero.png \
  --json

On success, stdout contains exactly one JSON document:

{
  "status": "completed",
  "result": {
    "jobId": "job-abc12345",
    "provider": "flux",
    "outputPath": "/abs/path/public/hero.png",
    "metadata": { "format": "png", "duration": 4321, "prompt": "..." }
  }
}

On failure: {"status":"failed","error":"..."} with exit code 1.

Human progress messages go to stderr. NO_COLOR env var disables colored output.

Other scripts

node scripts/doctor.js [--json] — report per-provider configured/unconfigured status. Run first if a generation fails with "not configured". JSON shape:

{
  "status": "ok",
  "providers": [
    { "name": "flux", "envVar": "REPLICATE_API_TOKEN", "configured": true, "source": "env" },
    { "name": "openai", "envVar": "OPENAI_API_KEY", "configured": false, "source": "unset" }
  ]
}

source is one of: agent-native | env | env-file | config-file | cli | unset.

node scripts/setup.js --provider <name> [--token <value>] [--persist env|config] [--force] [--json] Validate and persist a credential. Without --token, prompts on stderr. Validation calls the provider's cheapest authenticated endpoint; fails fast on 401/403. --persist config (default) writes to .nadicode-ai-designer/config.json (mode 0600). --persist env writes to ./.env. Both paths are gitignored.

node scripts/list-jobs.js [--status pending|running|completed|failed] [--json] List tracked jobs from .asset-jobs.json in cwd.

node scripts/validate-skill.js [--json] [--path <dir>] CI-time gate. Validates SKILL.md against Anthropic's quality rules (frontmatter limits, body length, reference depth, no Windows paths, no time-sensitive strings).

Secrets

Scripts resolve each provider credential in this order:

CLI --token flag  →  --env-file <path>  →  process env  →  .nadicode-ai-designer/config.json  →  defaults

Env vars:

Variable	Provider(s)
REPLICATE_API_TOKEN	Recraft + FLUX
OPENAI_API_KEY	OpenAI Images
IDEOGRAM_API_KEY	Ideogram
GOOGLE_API_KEY (fallback GEMINI_API_KEY)	Nano Banana + Veo

setup.js scrubs --token from argv and process.title immediately after parsing so the token does not linger in /proc/<pid>/cmdline or shell history.

Never commit tokens. See references/provider-knowledge.md for per-provider details and pricing.

Concurrency

Scripts persist to .asset-jobs.json in the current working directory with single-writer semantics — do not run multiple generate-* scripts against the same cwd in parallel. The jobs file uses atomic file replacement for writes, but the read-modify-write sequence is not locked across processes.