Skip to main content

Tableau Content Compass

Tableau Content Compass is a Codex skill for finding existing Tableau views, workbooks, datasources, projects, and Pulse metrics that match a business question. It uses Tableau MCP metadata plus a deterministic evidence-ranking helper to separate strong matches, partial coverage, and likely content gaps.

Codex conversion highlights​

  • Discovers the active Tableau MCP capabilities and schemas instead of assuming Claude-era tool names.
  • Supports scoped projects, project discovery, or bounded full-site search without always forcing an intake question.
  • Removes automatic memory reads and writes.
  • Keeps discovery read-only and separates it from workbook building, critique, documentation, and data querying.
  • Ranks normalized candidates with executable, tested logic.
  • Treats permission failures as unknown coverage rather than proof that content does not exist.
  • Uses only canonical identifiers and links returned by Tableau.

Installation​

Install the complete tableau-content-compass directory through your normal Codex skill or plugin workflow. Keep SKILL.md, agents/, and skill-expertise/ together.

Requirements​

  • Codex with the skill available.
  • A Tableau MCP server exposing one or more content-search or metadata capabilities.
  • Python 3.10+ for deterministic relevance ranking.

Example prompts​

Use $tableau-content-compass to find dashboards that show regional sales trends in the Commercial Analytics project.
Do we already have Tableau content for customer churn? Search views, datasources, and Pulse metrics.
Find the finance project first, then show me its strongest bookings and renewal content.

Ranking helper​

Normalize Tableau metadata using skill-expertise/ranking-schema.md. From the cloned package root, run:

cd tableau-content-compass
python3 skill-expertise/rank_content.py "/absolute/path/to/candidates.json"

The helper scores explicit metadata evidence, filters results below 15, groups matches into confidence tiers, and returns the two strongest signals for each explanation.

Confidence tiers​

ScoreTierInterpretation
60–100HighStrong evidence that the content addresses the question
35–59MediumPartial or adjacent coverage
15–34LowTangential but potentially useful
Below 15HiddenInsufficient evidence to recommend

Development​

Run tests and the active Codex skill validator before release:

python -m unittest discover -s skill-expertise -v

License​

No license has been assigned by this package. Add one before public redistribution.