Immediate result: A step-by-step cleaning plan with the code (pandas, SQL or spreadsheet steps) to do it.
Problem it solvesA messy export has duplicates, mixed formats and blanks, and cleaning it by hand is error-prone.
pandas · columns: name, phone, city, signup_date (formats: 03/10/26, 2026-10-03, '3 Oct')
signup_date: 3 formats → pd.to_datetime(df.signup_date, dayfirst=True, format='mixed')
city: 'Bombay'/'Mumbai '/'mumbai' → strip, title-case, map Bombay → Mumbai
Check: rows before = rows after + duplicates removed (logged).
A messy export with duplicates, mixed date formats and blanks is slow and risky to clean by hand. This skill profiles each column from a small anonymised sample, proposes a cleaning rule per issue and writes the code — pandas, SQL or spreadsheet steps. It never drops rows silently: everything removed is logged with a reason. It closes with validation checks such as row counts and totals that prove the cleaned data still adds up. You end up with a repeatable process rather than a one-off fix. To use it, paste the skill into your assistant and send column names and 10–20 sample rows (anonymised), plus the tool you use.
# Skill: Data cleaning plan
You are running the "Data cleaning plan" skill for me. I work as: Data analyst. What I need from you: a step-by-step cleaning plan with the code (pandas, SQL or spreadsheet steps) to do it.
## When I send you column names and 10–20 sample rows (anonymised), plus the tool I use
1. Profile each column: type, format issues, blanks, outliers, duplicates.
2. Propose a cleaning rule per issue.
3. Write the code or steps in the chosen tool.
4. Add checks that prove the cleaned data is right (row counts, totals).
## Reply in this format
- Column profile table
- Cleaning rules
- Code
- Validation checks
## Rules
- Never drop rows silently: log what's removed and why.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.
Confirm you're ready in one line, then wait for me to send column names and 10–20 sample rows (anonymised), plus the tool I use.Then sendColumn names and 10–20 sample rows (anonymised), plus the tool you use. For example: pandas · columns: name, phone, city, signup_date (formats: 03/10/26, 2026-10-03, '3 Oct')
Upload it to Claude, add it to ChatGPT's Codex app, clone it into Claude Code, or download the SKILL.md for any agent. Full guide with screenshots →
Download the ZIP and upload it in Claude. Claude then uses the skill on its own whenever a request matches.
Download ZIP
OpenAI's Codex app reads the same SKILL.md format. Copy the skill into your Codex skills folder:
$ git -c http.version=HTTP/1.1 clone https://scienceofapps.com/skills/ai-skills.git
$ cp -r ai-skills/data-cleaning-plan ~/.codex/skills/
Clone the library and copy this skill into your Claude Code skills folder.
$ git -c http.version=HTTP/1.1 clone https://scienceofapps.com/skills/ai-skills.git
$ cp -r ai-skills/data-cleaning-plan ~/.claude/skills/~/.claude/skills/ for all projects, or .claude/skills/ inside one project.Download the skill file and put it in a folder named after the skill.
Download SKILL.md~/.claude/skills/data-cleaning-plan/SKILL.md.Repositoryscienceofapps.com/skills/ai-skills.git · all 200 skills, one folder each, plain git over HTTPS
A messy export has duplicates, mixed formats and blanks, and cleaning it by hand is error-prone. The skill fixes that: a step-by-step cleaning plan with the code (pandas, SQL or spreadsheet steps) to do it.
Copy the skill and paste it as the first message of a new chat, then send your input. To keep it, paste it into a ChatGPT Project or custom GPT, a Claude Project or a Gemini Gem, or upload the SKILL.md file as a Claude Skill.
Send column names and 10–20 sample rows (anonymised), plus the tool you use. If something essential is missing, the skill asks you one short question and then continues.
Yes. It's free to copy, use and adapt, with no sign-up, and works in the free versions of ChatGPT, Claude, Gemini and Copilot.
ProblemDashboards get built with every metric and answer no question.
Weekly dashboard for the head of a food-delivery city team
Q1: Are we delivering on time? → On-time rate = orders delivered ≤ promised ETA ÷ delivered orders
Chart: line, 8 weeks, with target line at 92%
Filter: zone, hour of day.
Immediate resultA dashboard spec: the questions it answers, metrics with definitions, charts, filters and layout.
ProblemTwo teams report different numbers for 'active users' because nobody wrote the definition down.
'Activation rate' for a B2B SaaS — people mean different things
Activation rate = new workspaces that invite ≥ 1 teammate and create ≥ 1 project within 7 days ÷ new workspaces
Excludes: internal and test workspaces (domain list).
Edge: workspaces created on day 6 of the window are counted next week.
Immediate resultA precise metric definition: formula, inclusions, exclusions, grain, owner and SQL.
ProblemHundreds of open-text survey answers go unread, and only the scores get reported.
NPS survey, 340 answers to 'What's the main reason for your score?'
Slow delivery — 92 (27%) · 'Third late order this month'
Detractors mention delivery 3× more than promoters.
Action: show a realistic ETA at checkout, not the best case.
Immediate resultCoded themes with counts, representative quotes and what to do about each.
All 14 data & analytics skillsAll 200 skillsRead it in the articleProve it: AI for Data Analysts Certification (free test)
60 research-backed principles in 6 categories — what each one means, why it works on the human brain and how to use it on your own screens. Written by Sushant Singh, founder of Science of Apps, who has scaled consumer apps to lacs of users and spent 10+ years leading product across startups and digital businesses in the US, EU & India.
110 real paywalls from top apps, taken apart, rated out of 5 and decoded with the Laws of UX — plus the frameworks, copy and checklists to ship your own. By Sushant Singh, founder of Science of Apps.