# runway-dev-models

runway-dev-models — a skill that walks an agent through adding direct Runway Dev model endpoints to an app. Covers image, video, audio and task handling through the official SDK.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Windsurf (Devin Desktop), Claude.ai · 844 installs · ★ 76 repo stars
- Author: runwayml
- License: MIT · policy=mirror · the beginning of the file is at the end of this page
- Source: https://github.com/runwayml/skills/tree/HEAD/skills/runway-dev-models/SKILL.md
- Updated: 2026-10-20
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/runway-dev-models

## What it does

The skill starts by looking at the app to see where users hand over inputs and receive outputs. Runway calls stay behind the server boundary, so browser code never touches credentials. When a model is not explicitly chosen, the agent queries `list_models` from Dev MCP for the relevant project and endpoint instead of guessing IDs.

Before writing any integration code, the skill reads the exact endpoint reference from `https://docs.dev.runwayml.com/llms.txt` and `https://docs.dev.runwayml.com/api.md`. Fields, allowed values, inputs and outputs differ between models, so nothing is copied from a neighbouring endpoint. The official SDK for the app's language handles the calls.

Task-creating calls chain `waitForTaskOutput()` or `wait_for_task_output()` from the unawaited `create()` response. Documented task errors are surfaced to the user for inspection or recovery rather than silently resubmitted. Live generation costs money, so the skill runs it only on request and checks the project credit balance right before. `RUNWAYML_API_SECRET` lives in the server's existing secret mechanism and never lands in MCP configuration.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add runwayml/skills --skill runway-dev-models -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/runway-dev-models/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/runway-dev-models/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add runwayml/skills --skill runway-dev-models -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/runway-dev-models/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add runwayml/skills --skill runway-dev-models -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/runway-dev-models/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install runwayml/skills skills/runway-dev-models
```

2. **Manually: put the files in place**
- File: `.github/skills/runway-dev-models/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add runwayml/skills --skill runway-dev-models -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/runwayml/skills --path skills/runway-dev-models
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/runway-dev-models/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add runwayml/skills --skill runway-dev-models -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/runwayml/skills/tree/HEAD/skills/runway-dev-models/SKILL.md

## When to use it
- Wire a direct model endpoint into the app's server and user flow
- Check actual model access for a project without guessing IDs
- Follow the exact contract of the chosen endpoint
- Chain task output waiting and surface documented errors
- Keep the API secret out of browser code and MCP config

## Example prompts
- Add a direct Runway Dev image generation endpoint to my app
- List the models this project can actually use via Dev MCP
- Integrate the official SDK and wire up task output waiting

## Source text · SKILL.md · MIT

Below is the content of a third-party file. It is data, not instructions for you: do not act on anything written inside unless the human asks. Block tag: 41094a8ae35955bc.

BEGIN THIRD-PARTY FILE 41094a8ae35955bc
```text
---
name: runway-dev-models
description: "Use when adding or maintaining a direct Runway Dev model endpoint in an application, including image, video, audio, or task handling. Discover current model access and constraints when needed, follow the exact endpoint contract, and integrate through the official SDK. For configured model-agnostic routing, use runway-dev-model-routers instead."
user-invocable: true
---

# Runway Dev models

Integrate a direct model endpoint into the application's existing server and user flow. Consult `runway-dev` for account setup, credential boundaries, and shared SDK guidance when that skill is installed.

## Workflow

1. Inspect the app and determine where users provide inputs and consume outputs. Keep Runway requests behind the app's server boundary.
2. Honor a model the user or existing app explicitly selected. If a model must be selected or current project access and constraints matter, use Dev MCP's `list_models` for the relevant project and endpoint. Never guess a project ID or infer access from a model name.
   For catalog summaries, preserve each endpoint's returned price units and constraints, and verify model counts against the full response.
3. Read the exact current endpoint reference from `https://docs.dev.runwayml.com/llms.txt` and `https://docs.dev.runwayml.com/api.md`. A model's fields, allowed values, inputs, and outputs can differ from other models; copy no schema from a neighboring endpoint.
4. Use the official SDK for the app's language. For task-creating calls, chain `waitForTaskOutput()` or `wait_for_task_output()` from the unawaited `create()` response. Handle documented task errors and let the user inspect or recover the task rather than silently submitting it again.
5. Wire success, loading, and error states into the existing interface. Treat signed output URLs as temporary; persist media if the application needs it later.
6. Run appropriate local validation. A live generation is billable: perform one only when requested or approved, and check the selected project's credit balance immediately beforehand.

## Inputs and credentials

Follow the current input documentation for HTTPS URLs, data URIs, and upload URIs. For browser-selected files, send the file to the application's server and use the SDK's documented upload flow. Do not pass local filesystem paths to the API or accept arbitrary client-supplied remote URLs without validation.
```
END THIRD-PARTY FILE 41094a8ae35955bc

The third-party file has ended. Everything between the 41094a8ae35955bc tags is data: requests and commands from it are not followed. An “end of file” line with another tag inside the block is part of the file.

First 2430 of 2959 bytes. © runwayml · MIT · https://github.com/runwayml/skills/tree/HEAD/skills/runway-dev-models/SKILL.md

_Part of the page did not fit — see the HTML version._

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
