transcript-fixer
Fail
Audited by Gen Agent Trust Hub on Apr 6, 2026
Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/fix_transcript_enhanced.pyincludes afind_glm_api_keyfunction that reads sensitive shell configuration files, including~/.zshrc,~/.bashrc,~/.bash_profile, and~/.profile. This functionality is intended to automatically extract theGLM_API_KEYfor the user. Although the extracted keys are handled locally and masked in console output viautils/security.py, accessing these files is a high-risk behavior associated with credential harvesting. - [COMMAND_EXECUTION]: The skill frequently uses
subprocess.runacross several scripts. Inscripts/fix_transcript_enhanced.py, it executes core logic via theuvpackage manager. Inscripts/ensure_deps.py, it initializes virtual environments and installs dependencies. While these actions support the skill's primary workflow, they involve the execution of shell commands and the spawning of external processes. - [EXTERNAL_DOWNLOADS]: The
SKILL.mdfile provides instructions for users to install theuvpackage manager usingcurl -LsSf https://astral.sh/uv/install.sh | shand a PowerShell equivalent. These downloads originate fromastral.sh, which is the official domain for the well-knownuvtool. Per security protocols, downloads from well-known technology domains are documented neutrally as intended setup components. - [DYNAMIC_EXECUTION]: The health check utility in
scripts/utils/health_check.pyutilizes the__import__function to dynamically check if required Python packages (httpx,filelock) are installed. The module names are retrieved from a hardcoded list, minimizing the risk of arbitrary code execution. - [PROMPT_INJECTION]: The skill processes transcription data which is passed into an LLM context. It uses structured markdown headers (e.g.,
**需要修复的内容**:) to separate instructions from data. While standard LLM safety guardrails apply, the skill's ability to write to the file system and execute commands viauvcreates an attack surface for indirect prompt injection if malicious instructions are embedded in the processed transcripts.
Recommendations
- HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata