openless-voice-input
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a dictation pipeline that ingests untrusted audio data and processes it through an LLM before inserting the result into other applications.
- Ingestion points: Voice audio captured via the microphone in
recorder.rsand transcribed by external ASR services. - Boundary markers: The Rust backend includes system prompts (e.g., in
polish.rs) intended to restrict the LLM to cleaning text rather than answering questions. - Capability inventory: The application has the capability to insert text directly at the active cursor in any app using platform-specific Accessibility APIs or clipboard manipulation (
insertion.rs). - Sanitization: The skill instructions do not specify any validation or sanitization of the LLM-polished text before it is automatically inserted into external applications.
- [COMMAND_EXECUTION]: The "Build from Source" instructions provide standard developer shell commands including
npm ci,npm run tauri dev, and the execution of a custom build script./scripts/build-mac.sh. - [EXTERNAL_DOWNLOADS]: The documentation directs users to download pre-compiled application binaries (DMG and EXE files) from a GitHub repository (
github.com/appergb/openless).
Audit Metadata