speech-to-text
Fail
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- EXTERNAL_DOWNLOADS (HIGH): The skill's 'Quick Start' section recommends executing
curl -fsSL https://cli.inference.sh | sh. This is a highly insecure practice that executes remote code without prior verification or integrity checks. - REMOTE_CODE_EXECUTION (HIGH): Piped shell execution from an untrusted external domain allows the remote server to execute arbitrary commands on the user's local machine.
- COMMAND_EXECUTION (MEDIUM): The skill defines
allowed-tools: Bash(infsh *), granting the agent the ability to execute any subcommand of theinfshCLI, which was installed via an untrusted method. - CREDENTIALS_UNSAFE (LOW): The setup process includes
infsh login, which implies the storage and handling of authentication tokens for theinference.shplatform. While not hardcoded, the installation method makes these credentials vulnerable to the installed tool. - PROMPT_INJECTION (LOW): Category 8 (Indirect Prompt Injection) surface detected. The skill processes audio from untrusted external
audio_urlinputs. - Ingestion points: Audio data is fetched from external URLs provided in the input JSON (e.g.,
meeting.mp3,video.mp4). - Boundary markers: None identified in the provided
SKILL.mdto distinguish between transcription content and instructions. - Capability inventory: The skill has
Bashexecution capabilities via theinfshtool. - Sanitization: No evidence of sanitization or filtering for the transcribed text before it is returned to the agent context.
Recommendations
- HIGH: Downloads and executes remote code from: https://cli.inference.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata