读取和编辑 Word、PDF、Excel 文档。使用 python-docx、pdfplumber、openpyxl

This package appears to be a simple document reader, but the SKILL.md advertises a 'write' command that the included script does not implement — likely a documentation/code mismatch. Before installing or enabling: (1) review and test the provided doc_handler.py locally in a sandbox, (2) install the listed Python packages (python-docx, pdfplumber, openpyxl, pandas) in a controlled environment, (3) do not pass sensitive files to the skill unless you trust where its output will go (the tool prints file contents to stdout which could be forwarded by the agent), and (4) if you need write functionality, request the author/publisher for a corrected implementation or add/verify the write logic yourself. The mismatch suggests sloppy packaging rather than obvious malicious intent, but exercise caution and avoid granting the agent access to private files until you're comfortable with the code.

Type: OpenClaw Skill Name: doc-handler Version: 1.0.0 The skill is classified as suspicious due to a significant arbitrary file read vulnerability. The `doc_handler.py` script takes a file path directly from command-line arguments and attempts to read it, regardless of its location or sensitivity. This allows an attacker, through prompt injection against the AI agent, to potentially read any file on the system (e.g., `/etc/passwd`, `~/.ssh/id_rsa`). While the `SKILL.md` examples are benign, they demonstrate the agent's ability to execute arbitrary Python code, further highlighting the risk of code execution if the agent is prompted maliciously. There is no clear evidence of intentional malicious behavior like data exfiltration or persistence.