remove-ai-marks
Fail
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides setup scripts (setup_ctrlregen.sh, setup_markdiffusion.sh, setup_markllm.sh, setup_synthid.sh) that clone code from several non-trusted research repositories on GitHub (e.g., mertizci/noai-watermark, THU-BPM/MarkDiffusion, THU-BPM/MarkLLM, aloshdenny/reverse-SynthID). This code is subsequently imported and executed by the skill's Python backends, representing a risk of executing unverified code.
- [DYNAMIC_EXECUTION]: The skill uses eval() in scripts/classify.py and importlib.import_module() in multiple scripts to load components and optional backends dynamically from computed paths.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and analyze data from untrusted files and remote URLs (via audit_website.py, audit_dir.py, and inspect_file.py). This processed data is reported directly to the AI agent, creating a surface where a malicious actor could embed instructions in an asset's text or metadata to attempt to manipulate the agent.
- [COMMAND_EXECUTION]: The skill invokes various system tools and subprocesses, such as git, ffmpeg, exiftool, qpdf, and c2patool. It also includes logic in scripts/tui.py to replace its own process with bun.
- [EXTERNAL_DOWNLOADS]: The skill fetches configuration and guidelines from several well-known services and trusted organizations, including Hugging Face and Cloudflare. It also suggests installing bun using a piped shell command from its official domain.
Recommendations
- HIGH: Downloads and executes remote code from: https://bun.sh/install - DO NOT USE without thorough review
Audit Metadata