audio-transcriber

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill includes high-risk behaviors: automatic in-process package installation and explicit transmission of full transcripts/prompts to external LLM CLIs (claude / gh copilot), which can lead to unintended data exfiltration and supply-chain/environment modification risks.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.72). The required workflow sends the transcript text (derived from the user-provided audio content, i.e., potentially outsider-authored speech) and the user’s prompt into external LLM CLIs via subprocess.run(...) stdin/stdin-like input (Step “LLM Processing”), enabling indirect prompt injection through that free text.