Datacheck

Data Check

Inspect and validate a data file before analysis. Report findings with suggested fixes. Do not auto-fix.

When to Use

• Before any analysis or visualization on a new data file
• When parsing errors or unexpected values appear
• When switching to a different dataset mid-session
• User says "check", "inspect", "what's in this", "profile" a data file

When NOT to Use

• File has already been checked in this session and nothing changed
• User just wants to read a specific value (use Read tool directly)

Workflow

Step 1: Raw File Inspection

Before loading into R/Stata, check the raw file:

file                           # encoding detection
head -c 500  | cat -v          # hidden chars: ^M (CR), BOM, non-UTF8
wc -l                          # row count sanity check
head -3                        # peek at delimiter and header

Flag these issues: | Symptom | Meaning | |---------|---------| | ^M at line ends | Classic Mac \r or Windows \r\n line endings | | \xEF\xBB\xBF at start | UTF-8 BOM marker | | Only 1 line from wc -l | Entire file on one line (wrong line endings) | | Mixed delimiters | Inconsistent separator characters |

Step 2: Load and Summarize

R-first. Use Stata only when the file is .dta and project context is Stata-based.

For CSV files:

df 50 unique levels)
3. **Flag**: Numeric columns stored as character (likely parsing issue), zero-variance columns, duplicate rows

### Step 6: Report Summary

Present a concise summary with three sections:

**Clean:** Variables ready for analysis (no issues)

**Needs attention:** Variables with potential coded missings, encoding issues, or type problems. Include suggested fix code but do NOT execute.

```r
# SUGGESTED FIX for income_cat coded missings:
df$income_cat[df$income_cat %in% c(97, 98, 99)] <- NA

Notes: Observations about the data structure (survey weights detected, panel structure, etc.)

Quick Reference

| File type | Load with | Extra checks | |-----------|-----------|-------------| | .csv | read.csv() | encoding, delimiter, line endings | | .tsv | read.delim() | encoding, line endings | | .dta | haven::read_dta() | value labels, variable labels, weights, extended missings | | .rds | readRDS() | just structure check |

Common Mistakes

| Mistake | Fix | |---------|-----| | Loading .dta without haven | Always library(haven); read_dta(), never foreign::read.dta() | | Ignoring labelled class | Report label mappings; suggest as_factor() or manual recode | | Assuming 99 = missing | Check if 99 is in the real data range first | | Skipping raw byte check | ALWAYS run file and cat -v before loading. Mac CR issues are silent killers | | Auto-fixing without asking | Report + suggest. Never auto-recode missing values |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: letitbk
• Source: letitbk/claude-academic-setup
• License: MIT

Install and usage instructions live in the source repository linked above.