The Agent Skills Directory

[DATA_EXFILTRATION]: Path traversal vulnerability in scripts/generate_html.py. The script uses the process_images function to find image references in markdown and embed them as Base64 strings. It constructs the file path by joining the base directory with the image source (img_path = base_dir / src) without validating that the source remains within the intended directory. A malicious markdown file could reference sensitive system files (e.g., ../../.ssh/id_rsa), leading to their contents being exposed in the generated HTML article.
[EXTERNAL_DOWNLOADS]: Potential Zip Slip vulnerability in scripts/mineru_api.py. The script downloads ZIP archives from the MinerU API (mineru.net) and extracts them using zipfile.extractall(output_dir). This method does not prevent directory traversal if the ZIP file contains malicious filenames with relative paths (e.g., ../../target_file), which could be used to overwrite files outside of the output directory.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through the ingestion of external data.
Ingestion points: External PDF content is fetched and parsed via the MinerU Cloud API and local processing scripts.
Boundary markers: The skill does not implement delimiters or specific instructions to the agent to treat the parsed paper content as untrusted.
Capability inventory: The skill includes scripts with file system access (read/write) and network communication capabilities.
Sanitization: No validation or sanitization is performed on the text extracted from the academic papers before it is passed to the language model for synthesis and article generation.

paper-analyzer