tabular data processing and analysis

端到端表格数据引擎：提供工业级表格预处理（拆分、清洗、表头合并、描述生成），并支持在受控前提下进行深度探索性分析、可视化与报告生成。

This skill appears to be what it says: a Python-based table preprocessor and EDA helper that uses an OpenAI-compatible LLM for intelligent header merging and summaries. Before installing or running it, consider the following: - Treat OPENAI_API_KEY as sensitive: the skill will (by design) send headers/schema and sometimes sample rows to the configured LLM; do not enable LLM features on tables containing PII, financial data, credentials, or other sensitive content unless you accept remote exposure. Use the --no-abstract flag or disable merge_header when you want to avoid sending data to the remote LLM. - Dynamic code execution: the EDA step allows the agent to generate and run Python code. Run the skill in a sandboxed environment or isolated workspace, and review generated code before execution if possible to prevent accidental directory traversal or data exfiltration. - Use minimal-sending defaults: when you must use LLM features, prefer sending only schema, aggregated stats, or redacted samples. The skill's docs recommend these practices — enforce them in agent policies or by disabling LLM features when privacy is required. - Dependency installation: install requirements in a virtual environment and audit dependencies (openai, pandas, matplotlib). The repo does not perform external downloads beyond pip package installation. If you want a lower-risk setup: run the non-LLM parts (split/clean/transfer) locally and keep OPENAI_API_KEY unset; alternatively configure a private/enterprise LLM endpoint (OPENAI_BASE_URL) with strict monitoring and token scope limits. If you need additional assurance, request a code review focusing on any added/truncated files (the listing shows some files omitted) and enable runtime logging to observe which files are read and what is sent to the LLM.

Type: OpenClaw Skill Name: table-skill Version: 1.0.3 The table-skill bundle provides a comprehensive suite for industrial-grade table preprocessing, including splitting, cleaning, and metadata generation. While the skill instructions (SKILL.md and 05_eda_mining_skill.md) empower the AI agent to dynamically generate and execute Python code for Exploratory Data Analysis (EDA) and visualization, they include extensive and explicit safety guardrails. These instructions command the agent to follow the 'Minimum Necessary Principle,' avoid sending sensitive data (like PII or credentials) to LLMs, and restrict code execution strictly to data analysis tasks. The code utilizes standard libraries (pandas, matplotlib, openai) and lacks any indicators of malicious intent, such as data exfiltration, persistence mechanisms, or obfuscation.