nanoodle-skill

An Agent Skill that teaches any AI agent to build and run nanoodle graphs — multi-model AI media pipelines (text, LLM, image, video, audio) saved as a single noodle-graph.json and executed headlessly against the NanoGPT API with the user's own key.

The skill itself is SKILL.md. Once installed, an agent that supports the Agent Skills format (Claude Code, Cursor, Gemini CLI, OpenCode, and many others) will pull it in whenever a task looks like "generate an image/video/audio pipeline" or "run this nanoodle workflow".

This skill teaches your agent to design and run any graph — the right pick for one workflow, or for making new ones. Want prebuilt one-task skills instead (poster, jingle, video teaser)? → noodle-skills. Built up a whole folder of saved workflows to serve as typed, callable tools? → nanoodle-mcp. Running in GitHub CI? → run-noodle-action.

Using your own workflow

You don't install your graphs — you just hand one to your agent. No extra setup.

1. Build a workflow at nanoodle.com and press 💾 Save. You get a noodle-graph.json file, or a share link.
2. Ask your agent to run it — either works:
   • "Run poster.noodle-graph.json with Idea='a cozy ramen shop'"
   • "Run https://nanoodle.com/#g=…" (just paste the link)

The agent reads the workflow to find out what inputs it needs, then runs it. That's it.

Want a workflow to run automatically whenever you ask for that kind of thing (e.g. "make a poster")? Turn it into its own skill — see Making one workflow run automatically in SKILL.md.

Install

Via the open skills CLI (installs into ~/.agents/skills/ with vendor symlinks):

npx skills add nanoodlecom/nanoodle-skill -g -y

Or manually: copy this directory to .claude/skills/nanoodle-skill/ (project-level), ~/.claude/skills/nanoodle-skill/ (user-level), or wherever your agent reads skills from.

What's inside

nanoodle-skill/
├── SKILL.md                        # the skill: run graphs via CLI or JS/Python library
├── references/graph-format.md      # noodle-graph.json format + node-type table
└── examples/poster.noodle-graph.json  # runnable example: idea → LLM → poster image

Runs require a NanoGPT API key (NANOGPT_API_KEY) with balance — generations spend real money. Loading and inspecting graphs is free and offline.

Related

• nanoodle — the browser editor (nanoodle.com)
• nanoodle-js — nanoodle on npm: library + CLI
• nanoodle-py — nanoodle on PyPI: library + CLI

License

MIT — see LICENSE. Not affiliated with NanoGPT.