huashu-excel skill
Excel analysis skill that profiles a sheet, cleans it, reconciles totals against the sheet's own sums and then releases numbers.
- 464 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill is reference text and local Python scripts; the model attached to the agent sees the sheet contents.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/huashu-excelDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill is reference text and local Python scripts; the model attached to the agent sees the sheet contents.
This is third-party code. Review the repository files before installing.
What it does
The skill guides an agent through eight steps: profiling the sheet, cleaning, aligning on the question, analysis, reconciliation, delivery, chart verification and independent quality control. Profiling reads raw cells through `openpyxl` rather than `pandas` and finds two-level headers, merged cells, subtotal rows, duplicates and numbers stored as text. The skill uses total rows as checksums: it sums the cleaned data and compares. While `verify_numbers.py` finds a mismatch, the figures stay out of the report. Python scripts prepare tables and charts, and `scan_traps.py` looks for traps such as Simpson's paradox. The repository is also packaged as a DeepSeek Harness plugin.
Who it is for. Analysts and finance staff whose numbers from Excel exports must survive the question "how was this computed" and match the sheet's totals.
Good fit when
- An export with total and subtotal rows inflates the sum in `pandas`.
- Before a report you must confirm the sheet total matches its detail rows.
- A raw sheet must become an Excel or Word report with charts.
- A client asks why two figures disagree.
Not a fit when
- A quick look at data is enough: eight steps take time.
- There is no sheet and the data sits in a database; the skill targets Excel files.
- Machine learning models are needed rather than descriptive analytics.
Example request
Analyze this January sales sheet: find total rows and duplicates, reconcile the totals and only then state the overall sum.Limitations
The skill and README are written in Chinese, though answers can be in other languages. The Python scripts need `openpyxl`, and the agent takes the rest from its environment. The model reads figures from your sheets, so do not send confidential data to external services without permission. Independent QC needs a second agent. The agent runs the scripts itself, so read them before the first run. Results depend on the model, and conclusions drawn from data deserve a recheck.
How to disable. Delete the `huashu-excel` folder from the skills directory (`~/.claude/skills/` or `~/.agents/skills/`). The DeepSeek Harness plugin is removed with `dsh plugin --profile desktop remove huashu-excel`.
FAQ
Why not just drop the total rows?
The sheet's author computed them, so they work as checksums. If the cleaned detail does not sum to them, the error is in the detail or in the total.
What does `scan_traps.py` check?
Traps that reconciliation misses: Simpson's paradox and similar cases where numbers are right but the conclusion is wrong.
Why a five-number summary instead of the mean?
Business data is usually right-skewed, and the mean describes no typical object. The default output is minimum, quartiles, median and maximum.
Is DeepSeek Harness required?
No. The skill is shared across Claude Code, Cursor, Codex and other agents, and the DSH plugin is an extra option.
Which files does the skill produce?
Excel workbooks, charts and Word or HTML reports. The `verify_docx.py` and `verify_visual.py` scripts check the result before delivery.
Discussion
No questions or reviews yet.
Related
SkillsHigh-Stakes Analytics Lab
High-Stakes Analytics & Decision Lab
An evidence-gated analytics skill for high-stakes decisions: profiles data, picks a method the evidence supports, can answer with a data request
Skill and CLI for tables: cleaning with recorded decisions, level-based statistics, a PDF report and a separate audit.
A data visualization skill for agents: turns data into polished interactive HTML charts and full reports with one visual language
A skill that migrates a Tableau, Looker or Power BI dashboard to a Hex app and verifies numbers and looks with screenshots.
31 skills for Claude Code and Codex for the analyst workflow: data checks, cohorts, funnels, A/B tests, dashboards, summaries
A skill that teaches an agent to run Python scripts under noWorkflow, compare trials and trace where a result value came from.
README in short
The README shows a 161 percent error from ordinary sheet reading in a worked example, then describes the eight steps, five differences from other tools and a table of scripts.