Official Univer Skills for Spreadsheet Automation

Official Univer skills for spreadsheet automation across Claude Code, Codex, and Cursor.

This repository currently centers on one primary skill, agent-sheet: a workbook-native skill for agents that need real sheet structure, precise writes, import or export handoff, and verification-first spreadsheet workflows.

Why agent-sheet

agent-sheet is built for local, shell-native spreadsheet work:

• Fully local for core workbook workflows. No network connection or account signup is required.
• Powered by Univer, which has 12.6k GitHub stars as of March 20, 2026 and provides a modern spreadsheet engine without requiring Excel, LibreOffice, or a cloud spreadsheet account.
• Shell-native by design, so it works naturally with awk, grep, sed, jq, Python, and other native tooling.
• Better suited for large datasets and complex spreadsheet tasks where plain prompts or browser spreadsheets become brittle.
• Backed by the broader Univer spreadsheet engine ecosystem for rich workbook structure, formulas, sheets, ranges, and spreadsheet-native operations.

Quick Install

Recommended path:

# Install the workbook CLI used by the skill
npm install -g agent-sheet@latest

# Install this skill repository with the skills CLI
npx skills add dream-num/skills

Manual install:

git clone https://github.com/dream-num/skills.git
cd skills

# Claude Code
mkdir -p ~/.claude/skills
cp -R skills/agent-sheet ~/.claude/skills/

# Codex
mkdir -p ~/.codex/skills
cp -R skills/agent-sheet ~/.codex/skills/

# Cursor
mkdir -p ~/.cursor/skills
cp -R skills/agent-sheet ~/.cursor/skills/

Project-scoped installs follow the same pattern with .claude/skills/, .codex/skills/, or .cursor/skills/ in your repository root.

Available Skills

Skill	What it does	Best for	Status
agent-sheet	Spreadsheet automation with workbook inspection, precise writes, import/export handoff, and verification-first workflows	sheet inspection, formula review, bounded edits, review-table generation, shell roundtrips	primary

agent-sheet at a Glance

Use agent-sheet when the task needs:

• fully local spreadsheet work without network access or signup
• spreadsheet automation without depending on Excel, LibreOffice, or Google Sheets
• workbook inspection with explicit entryId context
• sheet, range, formula, and search analysis
• sparse writes, range replacement, anchored tables, and formula fills
• import or export handoff for local workbooks
• shell roundtrips with awk, grep, sed, jq, or Python followed by workbook-safe writeback and verification

Do not default to it for:

• plain CSV cleanup that does not need workbook semantics
• ad-hoc local workbook library code for operations already covered by the CLI
• mutation flows that skip verification after writing

Example Prompts

Use prompts like these after installing the skill:

Use agent-sheet to inspect this workbook, list all sheets, and summarize the formulas on the pricing sheet before making any edits.

Use agent-sheet to import the attached xlsx, add a review table to a bounded range on the QA sheet, then verify the anchor cells and header row.

Use agent-sheet to export the orders sheet, transform it with shell tools, write the cleaned result back as a table, and verify the structure plus the first rows.

Supported Clients and Requirements

Supported clients:

• Claude Code
• Codex
• Cursor

Runtime requirements from agent-sheet:

• OS: Linux or macOS
• Required binary: agent-sheet
• Common companion tools: awk, sed, python3 or python

Operational advantages:

• core workflows stay local
• no account registration required
• no Excel app, LibreOffice app, or cloud spreadsheet dependency
• native fit for shell pipelines and large data processing tasks

Install the CLI runtime with:

npm install -g agent-sheet@latest

Why Trust This Repo

• It is positioned as the official Univer skill repository for spreadsheet automation.
• It is powered by Univer, a widely used spreadsheet framework with 12.6k GitHub stars as of March 20, 2026.
• The repository is narrowly scoped around workbook workflows rather than generic prompt bundles.
• The primary skill favors explicit targeting, bounded writes, and post-mutation verification.
• Supporting assets are organized as readable playbooks, references, examples, and helper scripts instead of opaque generated blobs.

Safety Notes

• Review skills/agent-sheet/SKILL.md and any referenced scripts before installing in a trusted environment.
• Expect local command execution through the agent-sheet CLI and common shell tools.
• Treat write operations as incomplete until verification has been run on the relevant workbook surface.
• Prefer the smallest built-in write command that solves the task before falling back to custom scripting.

Validation Approach

This repository does not currently ship a separate evals/ directory. The current trust and validation signals come from the skill design itself:

• verification-first instructions in skills/agent-sheet/SKILL.md
• dedicated verification guidance in skills/agent-sheet/playbooks/15-verify.md
• concrete mutation and handoff examples in skills/agent-sheet/examples/
• helper scripts for spot-checking workbook outputs in skills/agent-sheet/scripts/

The roadmap for this repo includes stronger automated eval coverage.

Contributing

Contributions should preserve the current repo style:

• keep each skill self-contained under skills/<skill-name>/
• prefer explicit instructions over vague prompt prose
• keep reference material out of the main SKILL.md when progressive disclosure is clearer
• add verification guidance for any mutation-oriented workflow

License

This repository is licensed under the Apache-2.0 License.