Sherpa ONNX TTS

The stated purpose is local text-to-speech via sherpa-onnx, and the artifact's capabilities align with that purpose: downloading a runtime and voice model, setting local model/runtime environment variables, and invoking a TTS command.

The instructions are user-directed install and usage steps. They do not ask the agent to override user intent, access unrelated files, use credentials, or run background tasks.

Install metadata downloads and extracts native runtime/model archives from GitHub releases into local OpenClaw tool directories. That is coherent for offline TTS, but the artifact does not provide checksum or signature verification, and the referenced wrapper is not included in the reviewed files.

The requested environment variables are limited to local sherpa-onnx runtime and model directories. The optional PATH change is scoped to the skill's bin directory and no credentials or broad local indexing are requested.

Persistence is limited to downloaded tool/model files and user configuration in openclaw.json. There is no evidence of privilege escalation, startup persistence, destructive behavior, or long-running workers.