paper-summary-scripted

download arxiv paper pdfs or accept local paper files with a preprocessing script, then extract text, clean text, and generate a summary version, detailed ve...

This skill appears to do what it claims: it will download PDFs from arXiv (only) and read any local file paths you pass, extract text (requires pypdf for PDFs), write downloaded PDFs and extracted text into the output directory (./runs/paper-summary in the example), and then the agent will generate the three summarization stages plus a quality check. Before using: (1) ensure you only pass local file paths you trust (the script will open them); (2) be aware there is no builtin limit on downloaded file sizes or storage usage — provide an appropriate output directory and monitor disk space; (3) install pypdf if you need PDF extraction; (4) the prompts ask for citation-like support evidence (page numbers) but extracted plain text may not include page markers — expect that the contribution extraction step may need careful review. No credentials or external endpoints beyond arxiv.org are used.

Type: OpenClaw Skill Name: paper-summary-scripted Version: 1.0.0 The skill bundle contains vulnerabilities that could be exploited for shell injection and path traversal. In SKILL.md, user-provided inputs (paperurls and paperfiles) are wrapped in single quotes within a shell command, which is vulnerable to injection if the input contains unbalanced single quotes. Additionally, scripts/process_papers.py uses Path.resolve() on user-provided file paths without restricting access to a specific directory, potentially allowing the agent to read sensitive local files. While the arXiv downloader in the script correctly implements domain whitelisting (arxiv.org) to prevent SSRF, the lack of input sanitization in the command execution flow warrants a suspicious classification.