Normalize Tabular

Use this skill when the user wants to inspect, clean, standardize, or convert local

.csv

,

.xlsx

,

.xls

,

.h5

, or

.hdf5

files into a more analysis-friendly tabular format.

Prefer deterministic local conversion through the bundled script instead of doing large manual transformations in-model.

Bundled Tool

This skill expects the following helper script to exist:

• scripts/normalize_data.py

The script supports three commands:

• check-env

• inspect

• convert

Examples:

python3 "{baseDir}/scripts/normalize_data.py" check-env
python3 "{baseDir}/scripts/normalize_data.py" inspect "/path/to/data.csv"
python3 "{baseDir}/scripts/normalize_data.py" convert "/path/to/data.csv" --out "/path/to/out"
python3 "{baseDir}/scripts/normalize_data.py" convert "/path/to/data.xlsx" --sheet "Sheet1" --out "/path/to/out"
python3 "{baseDir}/scripts/normalize_data.py" convert "/path/to/data.h5" --dataset "pose" --out "/path/to/out"

Default Workflow

Follow this sequence unless the user explicitly wants only an environment check.

1. Check Environment First

Run:

python3 "{baseDir}/scripts/normalize_data.py" check-env

If required Python packages are missing, stop before conversion and explain the missing dependencies and install hint.

2. Validate The Input Path

Confirm the referenced local file exists.

3. Inspect First

Run:

python3 "{baseDir}/scripts/normalize_data.py" inspect "<input>"

Summarize the result for the user:

• detected source format
• rows and columns for CSV and Excel when available
• sheet names for Excel
• dataset names and shapes for HDF5

4. Decide Whether Conversion Is Safe

Convert immediately only when the source is unambiguous.

Unambiguous means:

• CSV: directly convertible
• Excel: only one sheet, or the user explicitly named the sheet
• HDF5: only one usable dataset, or the user explicitly named the dataset

5. Convert

When safe, run:

python3 "{baseDir}/scripts/normalize_data.py" convert "<input>" --out "<out_dir>"

If needed, include:

--sheet "<sheet_name>"
--dataset "<dataset_name>"

6. Report The Result

Always return:

• environment status
• what was detected
• what was converted
• output file paths
• notable layout or schema details
• warnings, fallbacks, or unsupported structures

Ambiguity Rules

Do not guess silently when structure is ambiguous.

Excel

If the workbook has multiple sheets:

• do not silently pick one
• list available sheets
• recommend the most likely rectangular data sheet if there is a clear candidate
• ask one focused follow-up only if needed

HDF5

If the file has multiple datasets:

• do not silently pick one
• list available datasets and shapes
• recommend the most likely tabular or pose dataset if there is a clear candidate
• ask one focused follow-up only if needed

HDF5 Rules

Never assume all HDF5 files are plain tables.

Possible structures include:

• 1D vectors
• 2D tables
• structured arrays
• pose tensors such as

  (frame, point, 2)

  or

  (frame, point, 3)

• higher-dimensional non-tabular arrays

Rules:

• only convert automatically when the script supports the detected structure safely
• never silently flatten unsupported higher-dimensional arrays
• report unsupported structures clearly

Output Contract

Prefer this output package:

<out_dir>/
  normalized/
    <name>.parquet
  schemas/
    <name>.schema.json
  reports/
    <name>.report.json

If parquet writing fails, a CSV fallback is acceptable.

Dependencies

Typical Python packages:

python3 -m pip install pandas openpyxl pyarrow h5py

OpenClaw Compatibility Notes

• Keep this skill self-contained
• Prefer the bundled script over ad hoc in-model table rewriting
• Keep YAML frontmatter minimal for compatibility
• If

  {baseDir}

  is not expanded by your runtime, replace it with the absolute path to this skill folder