voiceclaw

Voice conversation interface for OpenClaw using wake word detection, streaming LLM responses, and text-to-speech. Use when a user wants to talk to their Open...

This skill appears to implement the described voice → OpenClaw → VOICEVOX pipeline, but inspect and be aware of the following before installing: - The server auto-reads ~/.openclaw/openclaw.json to obtain an OpenClaw gateway token. Confirm you are comfortable with that file being read and that it does not contain secrets you don't want the skill to access. If you prefer, set OPENCLAW_GATEWAY_TOKEN explicitly in a controlled environment instead. - The registry metadata did not declare the config path or required env vars — treat that as an omission, not a guarantee of safety. Review src/server.js yourself (or run in an isolated VM/container) to confirm behavior. - SKILL.md and README advise cloning a GitHub repo and running npm install; ensure you clone the intended upstream repository and verify its integrity. Running code from GitHub and npm packages should be done from trusted sources. - The skill will make network calls to the configured OpenClaw gateway and VOICEVOX endpoints. Ensure those endpoints are local or trusted and that tokens used have limited privileges. - Recommended mitigations: run the skill in an isolated environment (container/VM), audit the bundled server.js for any unexpected data exfiltration, and explicitly set environment variables rather than relying on auto-discovery if you want stricter control. If you want, I can extract the exact lines where the code reads the home config and the env var usage so you can review them quickly.

Type: OpenClaw Skill Name: voiceclaw-jp Version: 0.1.0 The OpenClaw AgentSkills skill bundle 'voiceclaw-jp' is a voice conversation interface for OpenClaw. It reads the OpenClaw gateway token from `~/.openclaw/openclaw.json` and uses it to authenticate with the local OpenClaw Gateway. It also interacts with a local VOICEVOX TTS service. All file system and network access (to `127.0.0.1` by default) are directly related to its stated purpose. While user input passed to an LLM (via `/api/chat-stream` and `/api/chat` in `src/server.js`) inherently carries a prompt injection risk against the LLM, the skill itself does not craft malicious prompts or exhibit any intentional harmful behavior, data exfiltration, persistence mechanisms, or obfuscation. The `SKILL.md` and `README.md` provide setup instructions for the user, not malicious commands for an AI agent.