The Agent Skills Directory

Unverifiable Dependencies & Remote Code Execution (CRITICAL): The file docs/setup.md instructs users to install the uv tool by executing curl -LsSf https://astral.sh/uv/install.sh | sh. This pattern is a critical security risk as it executes unverified code from a remote server directly in the user's shell.\n- Unverifiable Dependencies & Remote Code Execution (HIGH): The skill dynamically downloads and executes model weights and code from Hugging Face (mlx-community/Kokoro-82M-bf16) and other sources (spaCy, espeak-ng) upon first run. These assets are not pinned by hash or verified, allowing for potential supply-chain attacks if the remote repositories are compromised.\n- Data Exposure & Exfiltration (LOW): The script scripts/generate_audio.py has broad file system access, allowing it to read any text file via the --file argument and write audio output to any user-specified path via --output.\n- Indirect Prompt Injection (LOW):\n
Ingestion points: Ingests untrusted text via CLI arguments or local files in scripts/generate_audio.py.\n
Boundary markers: Absent. The text is passed directly to the ML model.\n
Capability inventory: File system read/write, audio processing. No network exfiltration or command execution found in the core processing logic.\n
Sanitization: None. The skill assumes the input text is safe for processing.

local-tts