# Build with Runway Dev

Help the user add Runway Dev to the project in their current workspace. Their prompt may include a
model, endpoint, recipe, Character ID, router ID, or config ID; use that target without asking them to
choose it again.

1. Inspect the project quietly. Briefly explain what you will build, then integrate into the existing
   app. If the workspace is empty, ask what the user wants to build, recommend a web app as the
   starting point, and scaffold their choice.
2. Read https://docs.dev.runwayml.com/llms.txt and follow its task-specific links. Use the current
   docs for request fields, allowed values, models, and pricing. Do not guess API details.
3. If project-scoped skills are supported, install the shared Runway Dev skill and the relevant
   surface skill with:

   `npx skills add runwayml/skills --skill runway-dev --skill <surface-skill> --agent cursor claude-code codex -y`

   Choose `runway-dev-models`, `runway-dev-model-routers`, `runway-dev-characters`, or
   `runway-dev-recipes`. Scaffold first in an empty workspace. Read the installed `SKILL.md`
   files before continuing. If skills are unavailable, continue with the docs.
4. Probe for Runway Dev MCP tools quietly. Use them for live account state or approved account
   actions. If they are unavailable and live access is needed, follow
   https://docs.dev.runwayml.com/guides/mcp.md. If connection fails or the user declines, continue all
   account-independent work.
5. Implement the smallest useful end-to-end integration with the official Node or Python SDK. Most
   generation calls are asynchronous, so use the documented submit-and-wait flow. Only when making a
   live call, check for `RUNWAYML_API_SECRET`; if missing, ask the user to add it to the project's
   environment. Never request the secret in chat or hard-code it. Treat a billable test call as
   optional unless the user already requested one.

For an unspecified goal, recommend the shortest suitable surface: Models for direct generation,
Model Routers for automatic selection, Characters for realtime avatars, Recipes for packaged use
cases, or Workflows for multi-step pipelines. Ask before creating or changing live account state.

Finish the integration, run proportionate checks, and summarize what works and what still needs user
or account action.