Skip to main content

Codex Calc Engine

Codex Calc Engine helps Codex create, debug, translate, and optimize Tableau calculated fields. It combines field discovery, grain analysis, safe authoring, stored-formula read-back, and behavioral checks so a successful API response is not mistaken for a correct calculation.

When to use it​

  • Turn a business definition into a Tableau calculated field.
  • Repair mixed-aggregation, LOD, null, or table-calculation problems.
  • Translate supported SQL expressions into Tableau semantics.
  • Review an expensive calculation without changing its meaning.
  • Author a dependency-ordered batch of calculations.

Use a workbook-authoring skill for dashboard layout, visual styling, or broad .twb/.twbx changes.

Requirements​

  • Codex with this skill installed or available.
  • A Tableau Desktop agent or MCP runtime for live discovery, authoring, and behavioral verification.
  • Python 3.10+ for the optional local formula checker.

Without Tableau access, the skill can still draft formulas, but it will label storage and behavior as unverified.

Installation​

Install the entire codex-calc-engine directory through your normal Codex skill or plugin workflow. Keep SKILL.md, agents/, and skill-expertise/ together.

Example prompts​

Use $codex-calc-engine to create a 90-day retention rate using [User ID], [Order Date], and [First Order Date].
Debug this Tableau calculation. It returns “cannot mix aggregate and non-aggregate”: SUM([Sales]) / [Customer Target].
Translate this SQL window calculation into Tableau and explain the required table-calculation addressing.

Local formula checks​

Normalize a candidate formula for comparison:

python skill-expertise/calc_checks.py normalize --file formula.txt

Check a stored formula for common transport problems:

python skill-expertise/calc_checks.py check --file stored-formula.txt --expect-operator '<'

The checker is intentionally side-effect free. Tableau remains the authority for formula validity and workbook behavior.

Verification model​

The skill distinguishes three outcomes:

  1. Drafted — a formula was produced without live authoring.
  2. Stored — Tableau's stored definition was read back and matched.
  3. Behaviorally verified — representative workbook results were checked at the intended grain.

Blocked checks remain blocked; the skill does not infer success from missing runtime access.

Development​

Run the helper tests and 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.