Skip to main content

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​

PathPurpose
SKILL.mdCore routing, workflows, evidence, and authority boundaries
skill-expertise/tableau-evidence.mdMCP discovery, target resolution, evidence limits, and failures
skill-expertise/dashboard-documentation.mdBrief and detailed dashboard structures
skill-expertise/datasource-dictionary.mdField/calculation documentation guidance
skill-expertise/evidence-contract.mdEvidence-ledger schema
skill-expertise/output-formats.mdMarkdown, DOCX, XLSX, and update validation
skill-expertise/validate_evidence.pyDeterministic evidence-ledger validator
skill-expertise/test_validate_evidence.pySuccess, 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.