← Back to Skills Marketplace
Video Call AI Character
by
Yining Shi
· GitHub ↗
· v1.0.8
· MIT-0
340
Downloads
0
Stars
0
Active Installs
9
Versions
Install in OpenClaw
/install video-call-ai-character
Description
Video call the user with a real-time AI avatar powered by Runway. The agent initiates calls to the user — for standups, urgent alerts, check-ins, or any conv...
Usage Guidance
This skill is coherent with its stated purpose, but you should take the usual precautions before installing and running code fetched from npm and before providing an API key. Steps to reduce risk:
- Verify the npm package and repository: open the npm package page and the linked GitHub repository to confirm the package owner and that the source matches what you expect (look at recent commits, package.json, and the binary entrypoint).
- Inspect package contents before running (for example, review the repository or download the npm tarball to scan scripts). Avoid installing if the package owner is unknown or the code looks obfuscated.
- Treat RUNWAYML_API_SECRET like any API key: only provide it to services you trust. If possible, create a limited-scope key in Runway or rotate/delete the key after testing.
- Be aware the skill runs a local server and can optionally expose a tunnel URL. Only enable tunneling (cloudflared) if you understand the implications of exposing that endpoint and restrict who can access it.
- Decide whether you want the agent to autonomously initiate calls. If not, keep the skill user-invoked and confirm behavior in agent settings.
If you want a stronger assessment, provide the npm package tarball or the linked GitHub repository contents so the code can be inspected; that would increase confidence to 'high' if everything checks out.
Capability Analysis
Type: OpenClaw Skill
Name: video-call-ai-character
Version: 1.0.8
The skill facilitates AI-powered video calls by running a local Node.js server and optionally using `cloudflared` to create a network tunnel, which are high-risk behaviors even if plausibly required for the stated purpose. It instructs the agent to execute code via `npx openclaw-video-call` and manage media files in `/tmp`, while sending user and agent context to Runway ML (`dev.runwayml.com`) to generate avatars. No clear evidence of intentional malice was found, but the combination of remote code execution and network tunneling warrants a suspicious classification.
Capability Assessment
Purpose & Capability
Name/description (Runway-based real-time video avatar) match the declared requirements: node/npm and a Runway API key. The npm package openclaw-video-call and RUNWAYML_API_SECRET are directly relevant to the skill's purpose.
Instruction Scope
SKILL.md instructs running a local server (npx openclaw-video-call), creating avatars via the local API, and using the Runway service — all within scope. The instructions do not request arbitrary filesystem reads or unrelated credentials. They do recommend optionally installing cloudflared for remote access; that is clearly labelled optional.
Install Mechanism
Install is via an npm package (openclaw-video-call) which is a normal distribution channel for Node-based tools. npm packages are moderately risky compared to instruction-only skills because they run code fetched from a registry; the SKILL.md points to an npm page and a GitHub repo, but the package/owner should be verified before installing.
Credentials
Only one secret (RUNWAYML_API_SECRET) is required and that directly corresponds to use of Runway's API. No unrelated secrets, config paths, or excessive env vars are requested.
Persistence & Privilege
The skill does not force permanent inclusion (always:false). It runs a background server when the user starts it; because model invocation is allowed (platform default), the agent could autonomously initiate calls when permitted — this is expected for a call/alert skill but you should consider whether you want the agent to start outbound calls without manual confirmation.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install video-call-ai-character - After installation, invoke the skill by name or use
/video-call-ai-character - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.8
**Changelog for video-call-ai-character v1.0.8**
- Simplified initial avatar setup: the agent now builds personality without reading local identity files, using context it already has.
- Updated privacy notes: clarified that no local files are read, and added info on avatar deletion via API.
- Added repository link to metadata for easier source code access.
- Shortened and clarified instructions for creating avatars and setting up reference images.
- Removed references to optional reading of `IDENTITY.md`, `SOUL.md`, and `USER.md`.
- Updated documentation links and preset image recommendations.
v1.0.7
- Added a new Privacy & Data Handling section to clearly explain how avatar data, user information, and tunnels are managed.
- Clarified that identity files (`IDENTITY.md`, `SOUL.md`, `USER.md`) are optional and that providing the avatar personality can be done manually.
- Updated the "Building the avatar personality" section for better clarity on both manual and file-based setup.
- Adjusted metadata: removed `config` from the required agent setup, as identity files are now optional.
v1.0.6
Version 1.0.6
- Clarified options for the `referenceImage` field, listing HTTPS URL, Runway URI, and data URI, with examples.
- Added instructions for handling user-provided images, including options for small and large base64 images and how to upload large images to Runway.
- Updated the quick pick reference images list, including new "Openclaw Lobster" avatars at the top.
- Reordered reference image selection instructions to prioritize user-provided images before identity presets.
- Improved details for using pre-made character images and added text on preferred usage for large images.
v1.0.5
video-call-ai-character 1.0.5
- Made `referenceImage` a required field for avatars; you must now specify a face image when creating one.
- Clarified the documentation for custom avatar setup, including explicit instruction to use an avatar image from `IDENTITY.md`, the provided presets, or generate one.
- Updated the list and order of preset avatar images for easier selection.
- Rearranged and simplified API field requirements for creating avatars, clearly distinguishing required vs. optional fields.
v1.0.4
- Added explicit requirements for config files: IDENTITY.md, SOUL.md, and USER.md are now listed as required for setup.
- Included a "source" URL in the skill metadata for easier reference.
- No changes to core functionality or interface; documentation and metadata improvements only.
- Version update reflects enhanced clarity for setup and integration requirements.
v1.0.3
- Clarified avatar personality setup instructions to reference standard OpenClaw config files (`IDENTITY.md`, `SOUL.md`, `USER.md`), emphasizing these as the agent's own files.
- Streamlined and simplified the example personality prompt and instructions for building the `personality` field.
- Removed explicit references to agent abilities (like running code or managing files) from the example prompt, replacing them with a generic follow-up statement.
- Minor edits for brevity and clarity throughout the setup documentation.
- Fixed truncated pre-made character image URL and completed quick pick reference list.
v1.0.2
Version 1.0.2
- Updated avatar setup instructions with a new "Quick pick" option to easily choose from ready-made avatar images.
- Deprecated the section about generating custom images with Runway's text-to-image API for most users, streamlining the setup.
- Added URLs for various pre-made character images to simplify choosing a visual identity.
- No code changes; documentation improvements only.
v1.0.1
**Added workspace identity integration for avatar creation**
- Avatars now use data from your `IDENTITY.md`, `SOUL.md`, and `USER.md` to build more authentic personalities and visual identities.
- Instructions added for generating an avatar reference image using information from your identity files or Runway’s text-to-image if none exists.
- Expanded guidance on structuring the personality prompt to accurately reflect agent and user context.
- Voice selection guidance now matches available voices to your agent’s style from `SOUL.md`.
- Existing quickstart flow and API usage unchanged.
v1.0.0
Initial release of video-call-ai-character — AI video avatar for real-time calls.
- Enables an agent to initiate video calls with users using a Runway-powered AI avatar.
- Provides detailed setup instructions for creating a video call server and custom avatars.
- Supports delivering urgent alerts, standups, check-ins, and face-to-face conversations via a shareable call link.
- Offers transcript and recording retrieval for agent follow-up actions after calls.
- Includes avatar customization (name, voice, personality, and opening script) and context updates for each call.
Metadata
Frequently Asked Questions
What is Video Call AI Character?
Video call the user with a real-time AI avatar powered by Runway. The agent initiates calls to the user — for standups, urgent alerts, check-ins, or any conv... It is an AI Agent Skill for Claude Code / OpenClaw, with 340 downloads so far.
How do I install Video Call AI Character?
Run "/install video-call-ai-character" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Video Call AI Character free?
Yes, Video Call AI Character is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Video Call AI Character support?
Video Call AI Character is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Video Call AI Character?
It is built and maintained by Yining Shi (@yining1023); the current version is v1.0.8.
More Skills