The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of an external Python package using pip3 install evilarc. This tool is used to generate path traversal payloads in archives but represents an unverifiable third-party dependency.
[REMOTE_CODE_EXECUTION]: The documentation references a GitHub repository (https://github.com/mistymntncop/CVE-2023-4863) as a source for Proof-of-Concept code to target WebP processing vulnerabilities. This involves directing the user or agent toward untrusted external exploit code.
[COMMAND_EXECUTION]: The skill utilizes several shell commands and utilities including curl, pip3, python3, zip, tar, and cat for file creation and manipulation.
[DATA_EXFILTRATION]: The skill provides multiple payloads designed for data exfiltration and internal resource access:
Attempts to read sensitive system files such as /etc/passwd via symlinks and HTML IFrames.
Targets internal cloud metadata services at 169.254.169.254 to harvest security credentials via ImageMagick, FFmpeg, and Headless Chrome SSRF techniques.
Includes payloads to exfiltrate document content to an external host placeholder (COLLAB_HOST).
[DYNAMIC_EXECUTION]: The skill dynamically generates various exploit files (.mvg, .svg, .m3u8, .txt) locally using shell redirection (cat > /tmp/...) before processing or uploading them.
[INDIRECT_PROMPT_INJECTION]: The skill contains a significant attack surface for indirect injection:
Ingestion points: The skill creates and processes multiple external file formats (SVG, DOCX, ZIP, M3U8) which are then submitted to target endpoints.
Boundary markers: There are no markers or instructions to isolate or ignore embedded malicious instructions within these processed files.
Capability inventory: The skill possesses capabilities for package installation, file system writes, archive creation, and network requests.
Sanitization: No evidence of sanitization or validation for the content being processed was found.

hunt-file-upload