text-normalization-and-large-file-processing

The SenseNova model family plugs directly into agent runtimes such as OpenClaw and hermes-agent...

审核状态：已审核Quality 80Security 80

复制安装命令

用 Codex 或 Claude 安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它先审查 Skill 页面再帮你安装。

复制前请先查看来源、License 和安全提示。

项目 README

来源文件：README.md

抓取于 2026年8月3日

SenseNova-Skills

English | 简体中文

The SenseNova model family plugs directly into agent runtimes such as OpenClaw and hermes-agent, with the skills in this repository extending the models with concrete, end-to-end office capabilities.

In this repository each skill lives in its own directory and declares triggers, capabilities, and execution flow through a SKILL.md file, following the Agent Skills convention.

The skills cover image generation & visualization, slide-deck (PPT) generation, Excel data analysis, and deep research — usable standalone or composed into end-to-end workflows.

🎨 Want to see what it can do? Check out our sn-infographic Gallery to explore nearly 100 stunning generation cases and steal their prompt designs !

🦝 Available out-of-the-box in Raccoon

The latest SenseNova models and the full Cowork-Skill suite in this repo are bundled into Raccoon, with enterprise-grade security and a zero-setup experience — if you'd rather not provision env, API keys, and runtimes yourself, you can use these capabilities directly through Raccoon. Free trial available — no payment required to get started.

Raccoon now ships a full upgrade across product capability and client experience:

Three core office capabilities, strengthened: powered by SenseNova 6.7 Flash + Cowork-Skill, data analysis, PPT generation, and task planning each take a step up — covering the full loop from multi-file cleaning/analysis to formal report decks, industry/competitive research, and investment memos.
New: infographic generation: built on the SenseNova U1 model, compresses complex data, long reports, and business insights into dense, structured, visual infographics that are easier to digest and share.
New client + local Agent OS: the cloud model handles heavy reasoning and multimodal understanding; the local Agent OS sits next to your files, work context, and personal habits — delivering a more personalized, local, and secure AI-native office experience.
Proven at scale: chosen by 15M+ individual users and thousands of enterprise customers.

👉 Try it: xiaohuanxiong.com

How to Use

These skills are designed to run inside an Agent Skills-compatible agent.

Recommended runtime: pair them with OpenClaw or hermes-agent.
Recommended LLM: pair them with the SenseNova Platform API — a free token plan is available.
Install & configure: follow the full walkthrough in INSTALL.md.

Recommended: let the agent install the skills for you. Hand it the repo URL and ask it to clone and drop the skills into the right directory — for example:

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

After it finishes, you may need to manually restart the agent service before the new skills are picked up.

Agent	Target directory
OpenClaw	`~/.openclaw/skills/`
hermes-agent	`~/.hermes/skills/`

Prefer to install manually?

Clone this repository, then copy the subdirectories under skills/ into the target directory yourself:

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

For Hermes, swap the target to ~/.hermes/skills/.

Per-category Python dependencies, API keys, and invocation examples are documented in the 📖 Full guide for each section.

Skills List

🎨 Image & Visualization

📖 Full guide: docs/sn-image-generate_en.md (prerequisites, Quick Start, API config, and invocation samples).

Name	Label	Description
`sn-image-doctor`	Environment Doctor	Validates the SenseNova-Skills environment — checks `sn-image-base` install, Python deps, and required env vars; interactively fills missing values into `.env`.
`sn-image-base`	Image Base Layer (Tier 0)	Low-level tools — text-to-image (`sn-image-generate`), image recognition (`sn-image-recognize`), and text optimization (`sn-text-optimize`) — exposed through a unified `sn_agent_runner.py`, designed to be called by upper-layer skills.
`sn-infographic`	Infographic Generation (Tier 1)	Auto prompt-quality scoring, layout/style selection (87 layouts / 66 styles), multi-round generation with VLM review and quality ranking, producing publication-ready infographics.
`sn-image-imitate`	Image Imitation (Tier 1)	Given one reference image and a target content prompt, generates a new image that imitates the reference.
`sn-image-resume`	Resume Image Generation (Tier 1)	Given resume information, generates a resume image.

📊 Presentations (PPT)

📖 Full guide: docs/sn-ppt-generate.md (prerequisites, Quick Start, API config, and invocation samples).

Name	Label	Description
`sn-ppt-entry`	PPT Entry Point	Unified entry point for PPT generation. Asks the user to choose fast, standard, or creative mode, then collects role / audience / scenario / page count. For standard mode, also asks about image sourcing (AI, web search, or none) and chart rendering (U1 infographics or ECharts). Parses uploaded pdf / docx / md / txt, emits `task_pack.json` + `info_pack.json`, and dispatches to the chosen mode.
`sn-ppt-doctor`	PPT Environment Doctor	Environment check for the PPT pipeline — validates `sn-image-base`, API keys, the Node runtime, and optional deps; writes missing required vars into `.env`.
`sn-ppt-creative`	PPT Creative Mode	One full-page 16:9 PNG per slide, generated via `sn-image-generate` with a per-page composed prompt. Falls back to web image search when T2I generation fails.
`sn-ppt-standard`	PPT Standard & Fast	`style_spec` → outline → asset plan + per-slot images + VLM QC → per-page HTML → per-page review → PPTX export. Fast mode builds a complete draft immediately with autonomous decisions, then provides structured refinement suggestions. Supports AI-generated infographics (U1) for diagrams and web image search (Serper) for real photos.

📈 Data Analysis (DA)

📖 Full guide: docs/sn-data-analysis.md (prerequisites, Quick Start, API config, and invocation samples).

Name	Label	Description
`sn-da-excel-workflow`	Excel Analysis Orchestration	End-to-end Excel pipeline — multi-sheet read, large-file detection (≥10k rows triggers Parquet), cleaning, conditional filtering, cross-sheet aggregation, and Excel/CSV export.
`sn-da-image-caption`	Image Understanding & Data Extraction	For image-first inputs — table OCR, chart understanding, screenshot/UI description; parses captions into DataFrames, recreates visualizations, exports Excel/CSV.
`sn-da-large-file-analysis`	High-Performance Large-File Analysis	Streaming reads for ≥10k-row Excel datasets (openpyxl read_only + iter_rows), Parquet conversion, memory optimization, chunked processing, large-file writes.

🔬 Deep Research

📖 Full guide: docs/sn-deep-research.md (prerequisites, web_search precheck, Quick Start, and per-stage invocation).

Name	Label	Description
`sn-deep-research`	Deep Research Entry Point	Unified deep-research orchestrator with true-dependency DAGs, reusable source snapshots, and evidence-informed content units, producing final `report.md`.
`sn-research-report`	Final Report Writing & Editing	Renders the judgment layer into the final `report.md`; also handles targeted rewrites — restructuring, polishing, table-augmentation — for an existing draft.
`sn-report-format-discovery`	Presentation-Format Discovery	Compares final forms such as a research report, academic paper, table-first analysis, decision memo, or a custom Markdown form; scout uses it before research and user confirmation.
`sn-prepare-citations`	Citation Rendering	Post-processes `[^source_id]` footnotes into numbered citations and appends references from evidence sources.
`sn-md-to-html-report`	Markdown → HTML Report	Converts the research `report.md` (or any Markdown doc) into a clean, single-file HTML reading view that opens offline — embedded images, side-panel TOC, responsive tables, and table-delimiter repair.

🔍 Search

📖 Search skills are documented together with deep research: docs/sn-deep-research.md (includes per-platform API keys, invocation, and unified JSON output).

Name	Label	Description
`sn-search-academic`	Academic Search	ArXiv (with section-level HTML reading) / Semantic Scholar (with citation counts) / PubMed (with PMC open-access full text) / Wikipedia, in one aggregated interface.
`sn-search-code`	Developer Search	GitHub (repo / code / issue) / Stack Overflow / Hacker News / HuggingFace (models / datasets / spaces), aggregated.
`sn-search-social-cn`	Chinese Social Search	Bilibili / Zhihu / Douyin search; some platforms require cookie auth.
`sn-search-social-en`	English Social Search	Reddit / Twitter (X) / YouTube search.

Sample Outputs

🎨 Infographic (sn-infographic)

A few sn-infographic outputs (more in docs/sn-infographic-examples.md).

🧩 Memory price analysis — insight → analysis → presentation → end-to-end workflow

examples/memory-price-end2end-analysis. Starting from a raw quote CSV, the agent profiles fields, normalizes categories and timestamps, then attacks the rally from three angles — overall trend, top movers per category, and the gap between server-grade and consumer-grade SKUs — locating a late-February inflection along the way. Treating those findings as the research question, it switches to deep research: planning per-dimension web searches over supply contraction, AI-server demand, and vendor output discipline, then triaging and cross-checking evidence across sources before committing it to the report. The data and research conclusions are then handed to PPT generation, which lays out a 16-page outline, plans per-slot imagery, renders per-page HTML, runs VLM review, and finally composites screenshots into the PPTX. The result is a clear three-step storyline: prices are rising → here is why → here is what to do. This is the only example that exercises the full data analysis → deep research → PPT chain end-to-end.

Depends on: sn-da-excel-workflow, sn-deep-research, sn-ppt-entry, sn-ppt-standard, sn-md-to-html-report

📊 Employee performance analysis — data analysis

examples/employee-performance-analysis. The agent reads 10 separate monthly review xlsx files, aligns column schemas across months and joins them into one longitudinal table. From that table it produces aggregate views — monthly average trend, score-distribution boxplots, grade mix change, and a 38-role ranking — and individual views — top performers, needs-attention, and consistently-improving cohorts plus per-employee year trends. The findings are written up with explicit improvement suggestions tied to specific roles and individuals, backed by 8 supporting charts. The same content is delivered as a Word doc (for distribution) and a visualized HTML report (for browsing). The example shows how sn-da-excel-workflow handles "many small spreadsheets that should be one analysis" rather than a single big file.

Depends on: sn-da-excel-workflow

🔬 Embodied AI industry research — deep research

examples/embodied-ai-deep-research. Given only an industry name, the agent first commits to a research plan — market size, vendor share, financing, cost structure, development roadmap — instead of jumping straight into search. For each dimension it runs targeted web searches, fetches and reads source pages, and extracts both numeric and qualitative evidence; conflicting figures across sources are explicitly reconciled before being trusted. A synthesis stage organizes per-dimension evidence into a traceable, reader-oriented information structure rather than a stack of disconnected bullets. The output is an illustrated report (Markdown + visualized HTML) with 5 dimension-specific charts. The example shows how sn-deep-research turns "go research X" into a structured plan-then-execute loop with traceable evidence.

Depends on: sn-deep-research

🎯 Property fee pricing — PPT generation

examples/property-fee-pricing-ppt. The agent takes a free-form brief — topic (property fee pricing), audience (property staff + committee), 26 pages, black-and-white warm style — and first commits to an outline plus a per-page asset plan that conforms to the style spec. Each slide is then built as semantic per-page HTML rather than free-form image generation: copy, layout, illustrations, icons, and any data charts are reasoned about per slot. Imagery is produced or selected per slot and VLM-checked against the page's intent; each rendered page goes through a review pass with optional rewrite for coherence and copy quality. Final pages are screenshotted and composited into the PPTX, with the per-page HTML kept alongside for direct browser preview or re-editing. The example demonstrates sn-ppt-standard style consistency on a long, prose-heavy deck where every slide must obey the same audience and palette constraints.

Depends on: sn-ppt-entry, sn-ppt-standard

FAQ

Common setup and runtime questions (400/401 errors, rate limits, PPT timeouts, infographic quality, model names) are answered in docs/faq.md.

Contributing

Feel free to use the skills here as templates for your own OpenClaw skills. The qualities that make a skill good:

Clear triggers: state in description exactly when the skill should and should not run, so the agent recognizes it accurately
Focused scope: each skill does one thing well; complex workflows compose multiple skills
Solid documentation: examples, artifact contracts, edge cases, failure handling
Supporting resources: use references/, scripts/, prompts/ to provide additional context

Join the Community

Join our growing community to share feedback, get support, and stay updated on the latest developments. Scan the QR code below to hop into the chat — we'd love to hear from you!

Discord	Lark Group

License

MIT — see LICENSE.

其他

中风险

来源需自行核对维护者身份。
包含脚本或命令调用，安装前请复核。
未检测到明显外部权限要求。
未检测到高风险命令。
扫描发现：1 条。

Codex — Git Clone 安装

安装前请先查看来源仓库和风险报告。
克隆仓库：git clone https://github.com/OpenSenseNova/SenseNova-Skills.git
将 "skills/sn-da-excel-workflow/capability/excel-data-cleaning/text-normalization" 文件夹复制到 Codex 的 skills 目录中。
重启 Codex 让新的 skill 生效。

Codex — 手动复制安装

安装前请先查看来源仓库和风险报告。
从源仓库下载 SKILL.md 及相关文件。
在 Codex 的 skills 目录中创建新文件夹。
将所有 skill 文件复制到新文件夹中。
重启 Codex 让新的 skill 生效。

Claude Code — Git Clone 安装

安装前请先查看来源仓库和风险报告。
克隆仓库：git clone https://github.com/OpenSenseNova/SenseNova-Skills.git
将 "skills/sn-da-excel-workflow/capability/excel-data-cleaning/text-normalization" 文件夹复制到 Claude Code 的 skills 目录中。
重启 Claude Code 让新的 skill 生效。

Claude Code — 手动复制安装

安装前请先查看来源仓库和风险报告。
从源仓库下载 SKILL.md 及相关文件。
在 Claude Code 的 skills 目录中创建新文件夹。
将所有 skill 文件复制到新文件夹中。
重启 Claude Code 让新的 skill 生效。

Cursor — Git Clone 安装

安装前请先查看来源仓库和风险报告。
克隆仓库：git clone https://github.com/OpenSenseNova/SenseNova-Skills.git
将 "skills/sn-da-excel-workflow/capability/excel-data-cleaning/text-normalization" 文件夹复制到 Cursor 的 skills 目录中。
重启 Cursor 让新的 skill 生效。

Cursor — 手动复制安装

安装前请先查看来源仓库和风险报告。
从源仓库下载 SKILL.md 及相关文件。
在 Cursor 的 skills 目录中创建新文件夹。
将所有 skill 文件复制到新文件夹中。
重启 Cursor 让新的 skill 生效。

GitHub Copilot — Git Clone 安装

安装前请先查看来源仓库和风险报告。
克隆仓库：git clone https://github.com/OpenSenseNova/SenseNova-Skills.git
将 "skills/sn-da-excel-workflow/capability/excel-data-cleaning/text-normalization" 文件夹复制到 GitHub Copilot 的 skills 目录中。
重启 GitHub Copilot 让新的 skill 生效。

GitHub Copilot — 手动复制安装

安装前请先查看来源仓库和风险报告。
从源仓库下载 SKILL.md 及相关文件。
在 GitHub Copilot 的 skills 目录中创建新文件夹。
将所有 skill 文件复制到新文件夹中。
重启 GitHub Copilot 让新的 skill 生效。

Windsurf — Git Clone 安装

安装前请先查看来源仓库和风险报告。
克隆仓库：git clone https://github.com/OpenSenseNova/SenseNova-Skills.git
将 "skills/sn-da-excel-workflow/capability/excel-data-cleaning/text-normalization" 文件夹复制到 Windsurf 的 skills 目录中。
重启 Windsurf 让新的 skill 生效。

Windsurf — 手动复制安装

安装前请先查看来源仓库和风险报告。
从源仓库下载 SKILL.md 及相关文件。
在 Windsurf 的 skills 目录中创建新文件夹。
将所有 skill 文件复制到新文件夹中。
重启 Windsurf 让新的 skill 生效。

查看 SKILL.md 原文

name: text-normalization-and-large-file-processing
description: "对Excel文件进行文本标准化清洗（如去除异常前缀、提取纯中文字符等），并，最终输出清洗后的Excel文件并提供下载链接。"

Skill Steps

This sub-skill covers one capability of the Excel workflow. For reading/counting/Parquet optimization, see the parent workflow SKILL.md.

Step1 识别并清洗包含前缀符号的异常数值字段，统一转换为整数类型；同时使用正则表达式清洗文本字段，仅保留 Unicode 范围内的中文字符。

import re
import numpy as np

target_numeric_col = '需要转数字的文本列' # 示例：'获赞'
target_text_col = '需要提取中文的列' # 示例：'收货人'

# 1. 清洗包含前缀符号的数值字段
prefix_patterns = ['.', 'I ', '■ ', '一 ', '_', '. ']
def clean_numeric_with_prefix(value):
    val_str = str(value).strip()
    if val_str in ['None', 'nan', '', 'nan']:
        return np.nan
    for prefix in prefix_patterns:
        if val_str.startswith(prefix):
            val_str = val_str[len(prefix):].strip()
            break
    if val_str == '':
        return np.nan
    try:
        return int(val_str)
    except ValueError:
        return np.nan

# 2. 清洗文本字段，仅保留 Unicode 范围内的中文字符（\u4e00-\u9fff）
def clean_chinese_name(name):
    if pd.isna(name):
        return name
    s = str(name)
    chinese_chars = re.findall(r'[\u4e00-\u9fff]', s)
    cleaned = ''.join(chinese_chars)
    return cleaned if cleaned else ''

if target_numeric_col in df.columns:
    df[f'{target_numeric_col}_清洗后'] = df[target_numeric_col].apply(clean_numeric_with_prefix)
    
if target_text_col in df.columns:
    df[f'{target_text_col}_清洗后'] = df[target_text_col].apply(clean_chinese_name)

Step2 将清洗后的结果保存为 Excel 文件，在报告中提供下载链接，并执行内存清理以应对大文件处理时的内存压力。

output_path = '/mnt/data/标准化清洗结果.xlsx'

# 保存清洗结果
df.to_excel(output_path, index=False, engine='openpyxl')
print(f'清洗结果已保存到: {output_path}')

# 生成可下载链接
print(f'[下载清洗结果表](sandbox:{output_path})')

# 内存清理
if 'df' in locals():
    del df
    gc.collect()

发现问题？提交给管理员复核

暂无评论，成为第一个评论者吧！