For large models, being able to chat is no longer enough — the real bar is being able to deliver finished work: who runs the automated chain from a pile of dirty data to an analysis report to a presentation deck? SenseNova-Skills, open-sourced by SenseTime on GitHub, offers one answer. It is MIT-licensed and works across multiple agent frameworks including Claude Code, ChatGPT, and OpenClaw.

Before You Start

• An agent runtime (Claude Code / ChatGPT / OpenClaw / Hermes Agent all work)
• A SenseTime SenseNova API Key (optional; other models work too)
• GitHub repository: OpenSenseNova/SenseNova-Skills

What the Four Skill Categories Cover

1. Data Analysis (sn-da-excel-workflow)

Multi-sheet reading, Parquet optimization automatically triggered when a large file exceeds 10,000 rows, and cleaning, aggregation, and export — the entire pipeline is orchestrated. There is also sn-da-image-caption, dedicated to OCR for tables inside images.

Test scenario: storage chip price analysis — 7 months of quote records, 6 vendors and 13 categories mixed together, vendor names inconsistent in capitalization, nulls and outliers making up 13%.

The agent's approach:

• First runs a round of "data auditing" to fix inconsistent vendor naming, nulls, and other issues
• Analyzes the 13% outliers and chooses to keep and flag them (distinguishing real business situations such as channel inventory clearance and HBM spot-shortage premiums)
• Cross-analyzes by "category x application scenario x vendor" and accurately captures price inflection points
• Final conclusion: this round of price increases is a structural recovery driven by AI server demand, not a broad-based rally

2. Deep Research (sn-deep-research)

Contains 6 sub-skills covering planning, evidence gathering, synthesis, drafting, format discovery, and HTML conversion. Intermediate artifacts are persisted, so work can resume from breakpoints.

Test scenario: given only a topic, "low-altitude economy industry research", with no reference materials whatsoever.

The agent's approach:

• First anchors the core judgment (2025-2026 is the key commercialization window for China's low-altitude economy)
• Autonomously retrieves the latest progress from major domestic and international manufacturers, comparing aircraft models, airworthiness progress, and order data one by one
• Precisely extracts the UAM rollout pace and the localization rate of core eVTOL components
• Automatically generates an industry-chain structure diagram and a cost-share pie chart, and sizes the market with the TAM/SAM/SOM framework

3. PPT Generation (sn-ppt-entry)

Supports a standard mode and a creative mode. The standard-mode pipeline: outline -> per-page HTML -> per-page VLM review (auto-rewrite when a page fails) -> PPTX export.

Test scenario: "Layout and Operation Plan for Urban New Energy Vehicle Charging Infrastructure"

• Automatically builds a seven-part framework and delivers a complete 23-page PPT
• Matches each section with the right visual treatment (GIS heatmaps, comparison matrices, sensitivity analysis charts)
• The last page delivers implementation recommendations specific to responsible parties and timelines

Key point: the same skill can be dragged into ChatGPT and used there too.

4. Search

Aggregates arXiv, Semantic Scholar, PubMed, GitHub, Stack Overflow, Hacker News, HuggingFace, Reddit, Twitter, YouTube, Bilibili, Zhihu, and Douyin — full coverage of academic, developer, and Chinese/English social sources.

In addition, the skills can call the SenseNova U1 model to generate dense, complex infographics (automatic prompt expansion + VLM quality checks + quality ranking).

How to Install

Option 1: One-Click Install with Agent Pack

Project URL: OpenSenseNova/agent_pack

It bundles the Hermes Agent and OpenClaw frameworks plus the full skill set; combined with the free Token Plan, it runs as soon as it is installed.

Option 2: Manual Clone

git clone https://github.com/OpenSenseNova/SenseNova-Skills.git

Just copy the Skills directory into your own agent framework. You can even have the agent install it itself — tell it:

Please install https://github.com/OpenSenseNova/SenseNova-Skills into your Skills directory for me

It will clone, copy, and load everything on its own. The API is OpenAI-compatible, so mainstream open-source agent frameworks plug in smoothly.

Option 3: Cloud Product

SenseTime's office assistant Xiaohuanxiong (xiaohuanxiong.com) already bundles the entire skill line — ready to use out of the box, no code required.

Free Quota

SenseTime distributes a Token Plan through the SenseNova platform; in the first month of the promotion you get a quota of 1500 free calls every 5 hours.

Visit: sensenova.cn