SenseNova-Skills

What is SenseNova-Skills

SenseNova-Skills is SenseTime's open-source modular AI office skills library, designed specifically for Agent runtime. It breaks down four core office capabilities—infographic generation, PPT creation, Excel data analysis, and deep research—into independent, reusable skill modules that follow the agentskills.io specification and can be directly integrated into OpenClaw、Hermes Agent and other mainstream agent frameworks. At its foundation, it relies on SenseNova 6.7 Flash-Lite and U1 multimodal models working in division of labor, and through VLM quality closed-loop and streaming large-file processing technology, it achieves the ability to go from "providing an analysis approach" to "automatically running the complete workflow and directly delivering the finished product."

Main Features of SenseNova-Skills

• Infographic Generation: Supports 87 layouts and 66 styles, and through VLM automatic quality inspection and multiple rounds of iteration, transforms complex data, long reports, or business insights into high-density visual infographics.
• PPT Generation: Provides a standard mode (HTML pagination + VLM review → PPTX export) and a creative mode (generates a 16:9 U1 full-image PNG per page), and supports parsing PDF/DOCX/MD/TXT and other formats before automatically distributing tasks.
• Excel Data Analysis: Supports multi-table reading, automatic conversion of large files (≥10,000 rows) to Parquet, data cleaning, conditional filtering, and cross-table aggregation; hundred-thousand-row large tables use streaming processing to avoid memory overflow; supports OCR recognition of image tables and chart data extraction.
• Deep Research: Follows the standardized path of "planning → multi-dimensional targeted evidence gathering → cross-verification → comprehensive judgment," automatically identifies discrepancies in statistical calibers and contradictions in industry viewpoints, outputs traceable visual reports with rich charts, and supports resuming from checkpoints.
• Multi-source Search: Covers vertical search capabilities including academic (ArXiv, PubMed), developer (GitHub, HuggingFace), Chinese social (Bilibili, Zhihu, Douyin), and English social (Reddit, X, YouTube), etc.
• Modular Integration: Each skill is independently packaged as a directory + SKILL.md, and can either be called independently or combined into end-to-end complex workflows, directly connecting to Agent frameworks such as OpenClaw and Hermes Agent.

SenseNova-Skills Project Address

• Project Address：https://github.com/OpenSenseNova/SenseNova-Skills

How to Use SenseNova-Skills

• Agent Automatic Installation (Recommended): In an already deployed Agent (such as OpenClaw or Hermes Agent), send a natural language instruction, for example, "Please install SenseNova-Skills from https://github.com/OpenSenseNova/SenseNova-Skills to the skills directory," and the Agent will automatically clone the repository and complete the configuration. After installation, you may need to manually restart the Agent service.
• Manual Installation: Via git clone https://github.com/OpenSenseNova/SenseNova-Skills.git --depth=1 clone the repository, copy the subfolders under the skills/ directory to the corresponding Agent's skills directory (such as OpenClaw's ~/.openclaw/skills/ or Hermes's ~/.hermes/skills/), and according to INSTALL.md configure the API Key and dependencies.
• Office Raccoon (Zero-configuration Experience): Directly visit the Office Raccoon official website to use the intelligent office assistant, with no need to install, configure an environment, or set up an API Key, and freely experience the full set of SenseNova-Skills office capabilities.
• Configure API Key: After manual installation, you need to configure the SenseNova open platform API Key in the .env file (at https://platform.sensenova.cn/console/keys apply for free quota), and some search-type skills also require configuring separate API Keys for each platform.
• End-to-end workflow composition: A single skill can be invoked independently, or chained and combined as needed—for example, feeding data analysis results into deep research, then handing the research conclusions to a PPT generation skill, forming a complete "data → insight → report" workflow.

SenseNova-Skills' Core Advantages

• Modular and composable: Each skill is independently encapsulated as a directory + SKILL.md, following the agentskills.io specification, and can be invoked independently or flexibly chained into complex end-to-end workflows.
• End-to-end finished product delivery: It doesn't just output analysis suggestions; it automatically runs the entire process—data cleaning, chart generation, report writing, PPT composition, and more—directly delivering finished products ready for meetings and presentations.
• VLM quality closed loop: During infographic and PPT generation, a vision-language model is introduced for automatic quality inspection and multiple rounds of iteration, ensuring layout, color scheme, and image-text alignment meet publication-grade standards.
• Large-file streaming processing: For large Excel files (≥10,000 rows), it uses streaming reads + automatic Parquet conversion, stably handling data tables at the hundred-thousand-row level and avoiding memory overflow.
• Native multimodal architecture: The underlying layer relies on SenseNova 6.7 Flash-Lite and the U1 model working in division of labor—U1 handles image generation, while 6.7 handles planning, expansion, quality inspection, and text generation, achieving deep image-text collaboration.
• Standardized deep research path: It follows the methodology of "planning → multi-dimensional targeted evidence gathering → cross-verification → comprehensive judgment," automatically identifying discrepancies in statistical scopes and contradictions in viewpoints, and supporting resumption from checkpoints.

Comparison of SenseNova-Skills with Similar Competitors

Application scenarios of SenseNova-Skills

• Enterprise performance and HR data analysis: Batch process performance review Excel spreadsheets across months and multiple positions, automatically clean tens of thousands of data entries, generate complete Word reports and HTML visualization pages containing trend analysis, position comparisons, employee profiles, and improvement suggestions, directly usable for management meeting decision-making.
• In-depth industry research and investment assessment: With a one-sentence research request, automatically complete multi-source searches, cross-verification of evidence, and disambiguation of statistical definitions, and output a professional analysis report with more than ten charts and traceable sources, suitable for investors and industry researchers to quickly understand the current market landscape.
• Business presentation automation: According to the topic, audience, and style requirements, automatically plan the outline, allocate materials for each page, generate semantic HTML, and synthesize PPTX, suitable for long-text presentation scenarios such as product launches, project reports, and training courseware.
• Complex information visualization: Compress long reports, business insights, or technical concepts into high-density structured infographics, supporting 87 layouts and 66 styles, suitable for scenarios such as marketing materials, knowledge popularization, data journalism, and resume visual design.
• End-to-end data research chain: Starting from raw CSV/Excel data, first complete cleaning and trend analysis, then trigger in-depth research on anomalous data (such as "reasons for memory price increases"), and finally synthesize data conclusions and research insights into page-by-page professional PPTs, achieving full-chain automation from "data → insights → report"
• Academic and competitive intelligence gathering: Covering multi-source vertical searches such as ArXiv, PubMed, GitHub, HuggingFace, Zhihu, and Reddit, automatically integrating fragmented information into structured assessment conclusions, suitable for the daily intelligence work of researchers, product managers, and strategic analysis teams.