voice-skill

[Skill Scanner] Installation of third-party script detected This skill specification is functionally coherent for its stated purpose (voice conversations with Claude about local code) but contains privacy and supply-chain concerns: it requests and will transmit high-sensitivity data (arbitrary project files and git state) to external voice/AI providers, uses a public tunnel to expose a local server, and stores API credentials locally without detailed guidance about secure storage or scope-limiting file selection. These behaviors are not inherently malicious but are high-risk for data leakage if the implementation uploads snapshots or forwards files to third parties without explicit, fine-grained user consent. I rate this as suspicious/needs-review: the README is plausible, but the unclear data flows and use of tunnels/third-party providers warrant closer code-level review and runtime safeguards (least-privilege file access, encrypted credential storage, explicit upload consent, and clear provider endpoints). LLM verification: This skill's capabilities align with its stated purpose (voice-driven code review and live context). There is no clear evidence of intentionally malicious code in the provided documentation. However, the design grants broad read access to local project files and persists call transcripts, while routing code/context to external voice/LLM providers (Vapi, optional ElevenLabs). Those data flows are coherent with the feature set but represent significant confidentiality and supply-chain risks: unpin