Percept Voice Commands

Detects wake words in speech and routes voice commands like email, text, reminders, search, and notes to OpenClaw agents for execution.

This skill appears to do what it says (detect wake words and forward commands) but its instructions reference local files (percept/data/contacts.json and speakers.json), a dashboard (port 8960), a database, and an OpenClaw CLI without declaring required binaries, config paths, or any auth details. Before installing, ask the publisher for: (1) the exact CLI binary name and required invocation/authorization (how does agent authenticate to OpenClaw?); (2) explicit config paths and file permissions for percept/data/*.json and what data will be read or logged; (3) whether any credentials or network endpoints are required and how they are protected; and (4) an authoritative source or homepage (the SKILL.md links a GitHub repo but registry source is 'unknown'). Restrict the agent’s ability to execute high-risk actions (sending email/SMS, posting to calendars) or require explicit user confirmation for those actions until you’ve validated the integration. If you cannot obtain clear answers or verify the code/source, consider not installing or running this skill in a limited sandboxed environment.

Type: OpenClaw Skill Name: percept-voice-cmd Version: 1.0.0 The skill is suspicious due to a critical vulnerability described in `SKILL.md`. The documentation explicitly states that for 'General' commands, '[anything]' from the user's voice input will be 'forwarded to OpenClaw CLI for execution.' This instruction to the AI agent creates a severe prompt injection risk, potentially leading to arbitrary command execution (RCE) if the OpenClaw CLI or the underlying system does not adequately sanitize the user-provided input.