earllm-build
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill hardcodes a specific local filesystem path
C:\Users\renat\earbudllm. This exposes the author's local username and directory structure, which is a form of information disclosure. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to manage an Android application that transcribes voice input and sends it to an LLM.
- Ingestion points: Voice data captured from Bluetooth earbuds and transcribed to text via
SpeechToTextController.ktin thevoicemodule. - Boundary markers: There are no instructions for the agent or the application to use delimiters or ignore instructions embedded within the transcribed voice text.
- Capability inventory: The system has the ability to send data to external LLM APIs (
RealLlmClient.kt) and execute build/shell commands (gradlew,powershell). - Sanitization: The skill does not describe any sanitization or validation of the transcribed text before it is interpolated into the LLM prompt. An attacker could potentially use audio/voice input to inject instructions into the LLM pipeline.
Audit Metadata