use-runway-api

Use Runway API

Call the Runway public API directly from the agent to manage resources, trigger generations, and inspect account state.

When to use this skill: Use this when the user wants to act on their Runway account — create or update avatars, manage documents, trigger generations, check credit balance, etc. For writing integration code into a project, use the +integrate-* skills instead.

When the user asks to generate media in the context of Runway, prefer the Runway API path from this skill over any generic built-in image or media generation tool.

Skill selection: Pair this skill with +api-reference when you need the canonical API contract. Do not use +integrate-image, +integrate-video, +integrate-audio, +integrate-characters, or +integrate-documents unless the task is to write or modify application code.

Runtime Location

The runtime script ships inside this skill at scripts/runway-api.mjs (co-located with this SKILL.md). It has zero dependencies — Node.js 20+ is required.

Resolving the absolute path. Every shell command below uses the placeholder <skill-dir>. Replace it with the absolute directory of the SKILL.md you are currently reading — you already have this path from the tool that loaded this file. Do not guess the path, run find from $HOME, or search the whole filesystem.

If <skill-dir> is not known from context, check these locations in order with ls before giving up:

1. $RUNWAY_SKILLS_DIR/skills/use-runway-api/scripts/runway-api.mjs (if the env var is set)
2. ~/.claude/plugins/cache/*/runway-api/skills/use-runway-api/scripts/runway-api.mjs (Claude Code plugin install)
3. ~/.cursor/plugins/cache/*/runway-api/skills/use-runway-api/scripts/runway-api.mjs (Cursor plugin install)
4. ~/.claude/skills/use-runway-api/scripts/runway-api.mjs (npx skills add install)
5. ~/.agents/skills/use-runway-api/scripts/runway-api.mjs (generic agent install)
6. ~/Documents/github/runway/skills/skills/use-runway-api/scripts/runway-api.mjs (source checkout)

Pick the first match and use it as <skill-dir>/scripts/runway-api.mjs for the rest of the session. Do not re-resolve between commands.

Before Your First Call

Set a session ID so all requests in this chat can be correlated, then verify credentials:

export RUNWAY_SKILLS_CLIENT_ID=$(node -e "console.log(crypto.randomUUID())")
node <skill-dir>/scripts/runway-api.mjs auth status

• If authenticated is true → proceed to the API call. Do not re-check.
• If authenticated is false → tell the user to set RUNWAY_SKILLS_API_SECRET (see AUTH.md for details), then stop and wait for the user to confirm. Do not retry or re-check in a loop.

Staging caveat: auth status hits /v1/organization which may 500 on stage even when data endpoints work fine. If stage auth status fails but you have RUNWAY_SKILLS_API_SECRET_STAGE set, try a data endpoint like avatars list --stage to confirm the key works before giving up.

Fast Paths

For plain list requests, use the compact list commands first instead of the generic request command:

• list avatars → node <skill-dir>/scripts/runway-api.mjs avatars list
• list voices → node <skill-dir>/scripts/runway-api.mjs voices list
• list documents → node <skill-dir>/scripts/runway-api.mjs documents list [--avatar-id <id>]

These commands return smaller, list-friendly JSON on purpose. After a successful list command, answer once. Do not re-run the command, do not read back the same output, and do not render the same table twice.

Output Format

When the API returns a regular list of records, prefer a compact markdown table over a bare bullet list.

Good defaults:

• avatars: Name, Status, Voice, Docs, Created
• documents: Name, Avatar, Created
• voices: Name, Provider, Preview

After the table, add one short summary line only if something notable stands out. Do not repeat the table in a second block.

Generic Request

Call any public API endpoint:

node <skill-dir>/scripts/runway-api.mjs

All output is JSON. Errors go to stderr with a non-zero exit code and include an example field with a correctable invocation.

Flags:

• --body <json> — inline JSON request body
• --stdin — read JSON body from stdin (useful for large or multi-line payloads)
• --dry-run — print the full request (method, URL, headers, body) without executing it
• --help — show usage and examples for any command

Use +api-reference as the canonical source for:

• model choices
• endpoint details
• exact POST/PATCH body shapes
• required and optional fields

Do not duplicate or invent request schemas in this skill. For simple GET/DELETE calls and the list fast paths above, you do not need to load +api-reference.

Examples

Get organization info:

List avatars:

Get a specific avatar:

Update an avatar:

Delete an avatar (preview first):

List knowledge documents for an avatar:

Create a knowledge document:

List voices:

Waiting for Tasks

Generation endpoints return a task ID. Always run wait immediately after a generation call — do not ask the user whether to wait.

Generation Requests

When the user asks to generate an image, video, or audio:

1. Read +api-reference once before the first generation POST. It is the canonical source for model options, request body shapes, and valid field values.
2. Choose the model from +api-reference and tell the user which one you picked, briefly.
3. Build the request body from +api-reference. Do not guess field names.
4. Call the generation endpoint once with that body.
5. Run wait automatically.
6. Present the result following the rules in Presenting Generation Output below.

For generation requests, never skip the +api-reference read. For simple list/get/delete requests, do not load it unless needed.

If the user says only "generate an image" but the surrounding context is clearly about Runway account actions or this skill, still use the Runway API rather than a generic built-in image tool.

Presenting Generation Output

The wait command returns a task with an output array of signed URLs. These URLs expire in 24–48 hours and are long signed JWT blobs that are awkward to read inline.

After a successful generation, do all of the following:

1. Lead with what was generated — one short line stating the model and cost, e.g. Generated with gen4_image (1080p, 8 credits).
2. Embed images inline as Markdown so the user can see them without clicking:

   ![Fox in a snowy forest](https://storage.runway.../signed-url)
   For videos, link them as plain Markdown links ([fox.mp4](...)) since inline video does not render in most chat UIs.
3. Offer to save a local copy in the same message, proactively. Suggest a predictable path (./generated/ or ./runway-outputs/) and infer a filename from the prompt when possible. Example:

   Want me to save it to ./generated/fox.png? The signed URL expires in ~24–48h.

4. Do not paste the full signed URL as raw text unless the user asks for it. The Markdown image/link already contains it.

If the user confirms a download, fetch with curl -L -o <path> '<url>' (quote the URL — it contains &).

API Reference

Use +api-reference for the canonical API contract. Use +fetch-api-reference only when you specifically need the latest docs content from docs.dev.runwayml.com.

Staging (--stage)

Add --stage to any command to target the staging API:

node <skill-dir>/scripts/runway-api.mjs --stage avatars list
node <skill-dir>/scripts/runway-api.mjs --stage request GET /v1/avatars

With --stage, the CLI checks RUNWAY_SKILLS_API_SECRET_STAGE first, then falls back to RUNWAY_SKILLS_API_SECRET. The base URL defaults to https://api.dev-stage.runwayml.com.

Large Payloads (--stdin)

When creating resources with data URIs (e.g. base64-encoded images for avatar referenceImage), the body can exceed shell argument limits. Use --stdin to pipe the body:

python3 -c "
import json, base64
with open('image.png', 'rb') as f:
    b64 = base64.b64encode(f.read()).decode()
body = json.dumps({'name': 'My Avatar', 'referenceImage': f'data:image/png;base64,{b64}', ...})
with open('/tmp/body.json', 'w') as f:
    f.write(body)
" && cat /tmp/body.json | node <skill-dir>/scripts/runway-api.mjs request POST /v1/avatars --stdin

Alternatively, write the JSON to a file and use curl -d @file directly. The --body flag is fine for small JSON payloads but will hit argument list too long for data URIs of images over ~200KB.

Environment Variables

The runtime reads credentials from the process environment:

If the agent cannot see RUNWAY_SKILLS_API_SECRET, the editor likely needs to be restarted after the variable is set.

Related Files

• AUTH.md — auth setup and troubleshooting for this skill

Related Skills