Tableau Scribe
A Codex skill for producing evidence-grounded Tableau dashboard documentation and datasource dictionaries.
What it provides
- concise or detailed dashboard help content;
- business-friendly metric and field dictionaries;
- documentation updates against accessible current Tableau state;
- runtime discovery of Tableau MCP capabilities and identifiers;
- explicit separation of observed, inferred, and unknown claims;
- opt-in, bounded sample-data enrichment;
- validated Markdown, DOCX, or XLSX deliverables.
Example prompts
Document this Tableau dashboard for our sales managers.Create a brief help-page description for this view.Build a metric dictionary for measures and calculations in this datasource.Create an XLSX data dictionary for all accessible fields.Update this existing dashboard guide against the current Tableau view.Write these approved descriptions back to the specified Tableau fields if the runtime supports it.
Package layout
| Path | Purpose |
|---|---|
SKILL.md | Core routing, workflows, evidence, and authority boundaries |
skill-expertise/tableau-evidence.md | MCP discovery, target resolution, evidence limits, and failures |
skill-expertise/dashboard-documentation.md | Brief and detailed dashboard structures |
skill-expertise/datasource-dictionary.md | Field/calculation documentation guidance |
skill-expertise/evidence-contract.md | Evidence-ledger schema |
skill-expertise/output-formats.md | Markdown, DOCX, XLSX, and update validation |
skill-expertise/validate_evidence.py | Deterministic evidence-ledger validator |
skill-expertise/test_validate_evidence.py | Success, authorization, integrity, and portability tests |
Local validation
python -m unittest discover -s skill-expertise -v
python -m py_compile skill-expertise/validate_evidence.py
Validate a ledger from any working directory using absolute paths:
python /absolute/path/to/tableau-scribe/skill-expertise/validate_evidence.py /absolute/path/to/ledger.json --pretty
The skill does not query sample values automatically. Value enrichment requires explicit authorization and remains bounded to the documented purpose.