OpenSenseNova/SenseNova-Skills

SenseNova‑Skills – Office‑automation skills for AI agents

What it is – A collection of Agent Skills (self‑contained directories with a SKILL.md manifest) that extend the SenseNova family of large language models with end‑to‑end office capabilities. The skills can be dropped into any agent that follows the Agent Skills spec, such as OpenClaw or hermes‑agent, and then invoked through the SenseNova Platform API.

Why it matters – It turns a generic LLM into a full‑featured “AI coworker”: generate images, build infographics, analyse Excel workbooks, write research reports, and automatically produce PowerPoint decks. All of this is orchestrated by the agent runtime, so users only need to describe the high‑level goal (e.g., “make a 20‑slide deck on memory‑price trends”) and the skills handle data cleaning, web search, VLM review, and final file export.

---

Core skill groups

Category	Example skills	What they do
Image & Visualization	sn-image-base, sn-infographic, sn-image-imitate	Low‑level text‑to‑image, editing, recognition; higher‑level infographic generation with automatic layout/style selection and quality ranking.
Presentations (PPT)	sn-ppt-entry, sn-ppt-standard, sn-ppt-creative	Parse a brief, create an outline, plan assets, generate per‑slide HTML, run VLM quality checks, and export a PPTX. Supports fast, standard, and creative modes, plus optional AI‑generated charts or web‑sourced images.
Data Analysis	sn-da-excel-workflow, sn-da-image-caption, sn-da-large-file-analysis	Read multi‑sheet Excel/CSV, clean & aggregate data, stream large files (>10 k rows) with Parquet conversion, extract tables/charts from images, and output tidy Excel/CSV files.
Deep Research	sn-deep-research, sn-research-report, sn-prepare-citations	Multi‑stage research pipeline: plan queries, run parallel web searches (academic, code, social), synthesize evidence, generate a markdown report, format citations, and optionally convert to HTML.
Search	sn-search‑academic, sn-search‑code, sn-search‑social‑en/cn	Unified wrappers around arXiv, Semantic Scholar, PubMed, GitHub, Stack Overflow, Reddit, Twitter, Bilibili, etc., returning a normalized JSON payload for downstream skills.

---

How to get started

1. Choose a runtime – Install either OpenClaw or hermes‑agent (both support the Agent Skills spec).
2. Add the skills – The easiest way is to ask the agent to clone the repo and drop the skills into its skill directory, e.g.:

   "Please install SenseNova‑Skills from https://github.com/OpenSenseNova/SenseNova‑Skills into your skills directory."
   

   If you prefer manual steps:

   git clone https://github.com/OpenSenseNova/SenseNova-Skills.git --depth=1
   cp -r SenseNova-Skills/skills/* ~/.openclaw/skills/   # or ~/.hermes/skills/
   

3. Configure the SenseNova API – Obtain an API key from the SenseNova platform (international: https://platform.sensenova.ai, China: https://platform.sensenova.cn). Add the key and the correct base URL to a .env file; the *_doctor skills can help fill missing variables.
4. Run a skill – Each skill provides a short CLI entry point (e.g., sn-ppt-entry) that can be called directly or via the agent’s natural‑language interface.

---

Example end‑to‑end workflow

The repository ships a complete demo (examples/memory-price-end2end-analysis) that showcases the full pipeline:

1. Data analysis – sn-da-excel-workflow reads a raw CSV of memory‑price quotes, normalises timestamps, and produces cleaned Excel files.
2. Deep research – sn-deep-research plans and executes web searches on supply‑chain factors, aggregates evidence, and writes a markdown report.
3. Presentation – sn-ppt-entry + sn-ppt-standard turn the report into a 16‑page PPT, generating infographics where needed and performing VLM quality checks.
4. Export – The final PPTX and an HTML version of the report are written to the output directory.

---

Where to find more

• Docs & tutorials – docs/ contains step‑by‑step guides for each skill group (image generation, PPT, data analysis, deep research).
• Gallery – docs/sn-infographic-examples.md shows ~100 infographic prompts and results.
• FAQ – Common setup issues (auth errors, rate limits, timeouts) are answered in docs/faq.md.
• Raccoon UI – If you don’t want to manage runtimes or API keys, the same skill suite is bundled into the hosted Raccoon office assistant (free trial at office.xiaohuanxiong.com).

---

Who might use this

• Productivity‑focused AI agents that need to turn raw data or a simple user brief into polished deliverables.
• Enterprises looking to automate report generation, market research, or slide‑deck creation without building custom pipelines.
• Developers who want ready‑made skill templates to extend OpenClaw/hermes‑agent with their own multimodal capabilities.

---

Quick takeaways

• Modular – Each capability lives in its own skill directory with a declarative SKILL.md manifest.
• Model‑agnostic runtime – Works with any SenseNova model (e.g., 6.7 Flash, U1) accessed via the public API.
• End‑to‑end – From raw files → analysis → research → visualisation → final PPT/HTML, all orchestrated by the agent.
• Open source – MIT‑style repo; you can fork, modify, or contribute new skills.