data-describe

Data Describe

Generate AI-powered documentation for a tabular data file using describegpt. Produces a Data Dictionary (column labels, descriptions, types), a natural-language Description of the dataset, and semantic Tags — all via the connected LLM (no API key needed in MCP mode).

Cowork note: If relative paths don't resolve, call qsv_get_working_dir and qsv_set_working_dir to sync the working directory.

Steps

1. Index: Run qsv_index on the file for fast random access.

2. Profile: Run qsv_stats with cardinality: true, stats_jsonl: true to generate the stats cache. describegpt reads this cache for column metadata, so it must exist first.

3. Describe: Run qsv_describegpt with the requested options (recommend all: true for comprehensive output). At least one inference option (dictionary, description, tags, or all) is required. Output defaults to <filestem>.describegpt.md.

4. Present: Display the generated Data Dictionary table, Description, and Tags to the user.

Options

Option	Effect
--all (recommended)	Generate Dictionary + Description + Tags in one pass
--dictionary	Data Dictionary only — column labels, descriptions, types
--description	Natural-language dataset Description only
--tags	Semantic Tags only
--format	Output format: Markdown (default), JSON, TSV, TOON
--language	Generate output in a non-English language (e.g. Spanish, French)
--addl-cols-list	Enrich the dictionary with extra columns (e.g. "everything", "moar!")
--tag-vocab	Constrain tags to a controlled vocabulary (comma-separated)
--num-tags	Number of tags to generate (default: 5)
--num-examples	Number of example values per column in the dictionary
--enum-threshold	Max cardinality to treat a column as an enum in the dictionary

Notes

• No API key needed in MCP mode — uses the connected LLM automatically via MCP sampling
• The stats cache must exist first for best results (step 2 creates it)
• Output defaults to <filestem>.describegpt.md
• For Excel/JSONL files, the MCP server auto-converts to CSV first
• Use --format JSON when you need machine-readable output for downstream processing
• Use --language to generate documentation in the user's preferred language