Skillv1.0.2

ClawScan security

Her Voice · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignFeb 15, 2026, 10:19 PM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill implements an on-device TTS/visualizer and the files, commands, and configuration it requests are consistent with that purpose — nothing in the package looks designed to do something else.
Guidance: What to consider before installing: - This skill installs Python virtual environments and pip packages (mlx-audio, kokoro, numpy, etc.) and may download a TTS model (large). Ensure you have disk space and are comfortable allowing those network downloads. - On macOS it will attempt to compile a Swift visualizer and may run Homebrew to install espeak-ng; follow prompts and grant only the actions you trust. - The setup may patch a third‑party module inside the created venv to fix library loading (macOS only). The change targets the venv, not system packages, but you may want to inspect the patch before allowing it. - The optional daemon listens on a UNIX socket under ~/.her-voice with restrictive permissions (0600). Any local process with access to your user account could connect; the socket is not network-exposed. - No API keys or external endpoints are hardcoded; however pip/model downloads require internet access. If you need to be cautious, run setup in an isolated environment (VM/container) or review the scripts first. - If you are not comfortable with code that installs packages, writes to your home directory, compiles binaries, or modifies venv-installed modules, do not install or audit the code before running.
Findings: [NO_SCAN_FINDINGS] expected: Static pre-scan reported no injection signals. Manual review shows code modifies venv files (macOS espeak patch), installs packages via pip, creates venvs, compiles a Swift binary, and runs a local UNIX socket daemon — all expected for an on-device TTS skill.

Review Dimensions

Purpose & Capability: okName/description (local TTS and visualizer) align with the included files and runtime steps. The package reasonably needs Python, espeak-ng, venvs, optional Swift compilation, and model files; those are present and used for Kokoro/MLX TTS and the macOS visualizer.
Instruction Scope: noteThe SKILL.md and scripts direct the agent to run setup.py, start/stop a local daemon, run speak.py, and optionally compile/run a Swift visualizer. These instructions are within the stated TTS scope, but the setup step will install packages, create venvs, download models, and — on macOS — patch a third‑party module in the venv to fix library loading. Those actions are functional for the feature but worth the user's awareness.
Install Mechanism: noteThere is no registry 'install' spec, but setup.py will create virtual environments and invoke pip to install mlx-audio/kokoro and dependencies, and on macOS will call swiftc to compile the visualizer and may run Homebrew. This involves network downloads from PyPI and model sources (expected for TTS), which is moderate risk but proportionate to the feature.
Credentials: okThe skill requests no environment variables or external API keys. Configuration and credentials are stored under ~/.her-voice. The declared binaries (python3, espeak-ng) match the functionality and no unrelated credentials or system paths are requested.
Persistence & Privilege: okThe daemon is optional (not always:true). It creates confined files in ~/.her-voice, a UNIX socket and PID file with restrictive permissions (0600), and does not modify other skills or system-wide agent settings. Running a background daemon is expected for low-latency TTS.