Security audit
Security checks for vulnerabilities and agentic risk
This is a TTS voice-control skill with documentation gaps, but I found no hidden code, credential collection, exfiltration, or unsafe automatic behavior.
Install only if you are comfortable with a TTS skill that may use command execution, download voice models, store local TTS preferences/cache, and delete cached audio when cleanup is invoked. Review any update action before allowing it, because the artifact does not specify where updates come from or how they are verified.
The overview at L021 presents the skill as 'Free, offline, no account required,' implying offline operation. But L283 declares 'LLM API' as required, and L328 includes network failure handling, which indicates dependence on online/API-backed execution. This is an intent-level contradiction in the documentation about whether the skill actually works offline.
The skill advertises an update command without warning that it may change local files, dependencies, or behavior. In an agent environment with exec capability, users may invoke updates without understanding that this can modify the local installation or pull unreviewed content, increasing supply-chain and integrity risk.
The cleanup command is described without clearly warning that cached audio files will be deleted. In a skill with exec capability, ambiguous destructive commands can cause unintended data loss, especially if users rely on cached outputs for recovery, auditing, or replay.
At L285-L286, the documentation says the skill is Markdown-instruction based and requires no additional API key. However, the known limitations section at L343 says '需要API Key, 无Key环境无法使用' ('API Key required; cannot be used without a key'). These statements directly contradict each other about a material usage prerequisite.
The markdown explicitly lists activation keywords as "switch, code, ai-assistant, openclaw, agentvibes, skill," several of which are common or generic terms rather than narrowly scoped invocation phrases. Without clear constraints or exclusion examples, this can cause the skill to match ordinary conversation or unrelated requests.
Mentioning downloads from HuggingFace without a privacy/network disclosure is risky because users may assume the skill is fully offline while it can contact external services and fetch models. This can expose IP metadata, create compliance issues in restricted environments, and introduce supply-chain concerns if remote artifacts are not verified.
No suspicious patterns detected.
