| name | csv-profiler |
| description | Profile a CSV file with the Python standard library only — row and column counts, per-column type inference, null counts, and basic numeric statistics. Use when someone attaches or points at a CSV and wants to know what is in it before analysing it, or when pandas is unavailable or too heavy for the question. |
CSV Profiler
Answers "what is actually in this file?" before anyone writes an analysis
against it. Standard library only, so it runs in any sandbox — including one
built from a slimmer image than this bundle's.
When to use it
- A CSV arrives and you need shape, types, and null density before deciding
anything.
- A pandas read is failing and you need to know whether the file is the problem.
- The question is small enough that importing pandas costs more than it saves.
Use it
python3 skills/csv-profiler/scripts/profile_csv.py <path-to-csv> [--sample 5]
It prints a per-column table: inferred type (integer / float / date / boolean /
string), null count and percentage, distinct count, and for numeric columns
min / max / mean / median. --sample N adds the first N rows.
Then
- Read the null column first. A column that is 60% empty is a different
analysis than one that is 2% empty, and it is the most common reason a
later aggregate is quietly wrong.
- Check the inferred types against what the column is called. An
amount
column inferred as string means the file has currency symbols, thousands
separators, or a stray header row inside the data.
- Say the shape out loud in your answer — row count, column count, the
columns you are going to use — before you compute anything on it.
Do not report a statistic the profiler did not produce. If the question needs
something beyond the profile, load the file properly with run_python and
compute it; the profiler is a first look, not the analysis.
If the script cannot be read
On the Kubernetes path this file and the script reach the sandbox only through
the sandbox image. If python3 skills/csv-profiler/scripts/profile_csv.py
reports the file does not exist, do not reimplement it inline and do not guess
at its output — say the skill's script is not available in this deployment and
compute what you need directly with run_python. See the README's
skills_builtin section for the cause.
