Tableau Intake

tableau-intake

Step 2 of 8. Converts a free-form request into a structured PRD.md so the downstream planning steps work from an unambiguous spec. The step is optional (it can be skipped) and idempotent (re-running refines an existing PRD instead of clobbering it).

| | | |---|---| | Reads | The request: production DASHBOARD-REQUEST.md (preferred) or text the analyst pastes or the scaffold/EXAMPLE-DASHBOARD-REQUEST.md demo example. None is a required read — the request can be pasted, so this step never refuses for a "missing" request file (CONTRACT.md §1). | | Writes | PRD.md at the project root (latest approved truth; overwritten in place). | | STATE.md update | Sets intake = approved (PRD authored) or skipped; flips every downstream approved step to stale on a re-run (CONTRACT.md §4.2). | | Entry gate | Refuses to run until init is approved in STATE.md (CONTRACT.md §4.1). | | Next step | tableau-data (or tableau-route to confirm). |

Why optional? PRD.md is a structuring convenience, not a hard dependency. tableau-plan (step 5) reads it as an optional input and falls back to the raw DASHBOARD-REQUEST.md when intake was skipped (CONTRACT.md §1), so skipping never blocks the pipeline — it just means planning works from the unstructured request.

The mechanical guarantees — the entry gate, the PRD schema check, and the STATE.md transition — live in intake.py, this skill's executable mirror of the contract. Your job is the part that needs judgment: reading the free-form request and authoring (or refining) the PRD prose. Run the script at the two points below; do not hand-edit STATE.md yourself.

How to run

1. Precheck. From the project directory, run:

``bash python "${CLAUDE_SKILL_DIR}/scripts/intake.py" precheck "" ``

(Use python3 if python is unavailable.) If it prints [BLOCKED], relay the reason and stop — the analyst must run tableau-init first. Otherwise note its three signals: whether a PRD.md already exists, which request source to read, and the current intake status (a re-run).

1. Refine vs. overwrite. If precheck reports a PRD.md already exists, use

AskUserQuestion to offer Refine (keep its structure, update the content), Overwrite (author fresh), or Skip. Never silently overwrite an existing PRD.md. If no PRD exists, author fresh (still offer Skip).

1. Read the request. Follow precheck's request source:

• DASHBOARD-REQUEST.md — read the analyst's production request.
• scaffold/EXAMPLE-DASHBOARD-REQUEST.md — the demo fallback; **tell the

analyst** you're demoing the workflow, not using real input.

• none — ask the analyst to paste their request text.

1. Author PRD.md. Write (or, when refining, Edit) PRD.md at the project

root using references/PRD-TEMPLATE.md as the structure. Always include the required core — Overview and Visualizations. Proactively propose KPIs, Filters, and Additional Notes, since most dashboards want them — but if the analyst says a section isn't needed, omit it without friction. Don't complicate the PRD with empty sections. When refining, preserve the analyst's existing structure and custom sections; change only what the request updates. Present the PRD for approval.

1. Commit — only after the analyst approves (or chooses to skip):

``bash python "${CLAUDE_SKILL_DIR}/scripts/intake.py" commit "" --status approved # or, to skip the step: python "${CLAUDE_SKILL_DIR}/scripts/intake.py" commit "" --status skipped ``

On --status approved the script validates PRD.md has the required core; if it prints [REFUSED] listing missing sections, add them and re-run. On success it records the status and reports any downstream steps it marked stale. Relay the summary and tell the analyst to open a fresh conversation and run the next step (tableau-data, or tableau-route to confirm).

The PRD schema

intake.py enforces only the required core; the rest is recommended-but-optional.

| section | required? | what goes there | |---------|-----------|-----------------| | ## Overview | required | Purpose, audience, update frequency. | | ## Visualizations | required | The charts and what each shows. | | ## KPIs | optional | Headline metrics with formulas; omit if the dashboard has none. | | ## Filters | optional | Slicing controls; omit if users don't need filtering. | | ## Additional Notes | optional | Branding, conditional formatting, sort orders, constraints. |

The validator matches headings as case-insensitive substrings and allows extra custom sections, so a refined PRD that keeps its own structure still passes.

Notes

• Idempotent & non-blocking. Skipping records skipped in STATE.md and never

blocks the pipeline — downstream plan falls back to DASHBOARD-REQUEST.md and neutral handling (CONTRACT.md §1). Re-running refines the root PRD.md in place and flips downstream approved steps to stale; it does not create a new version directory (PRD.md is a root "latest truth" file, CONTRACT.md §4.3).

• Do the data/branding work elsewhere. Intake only structures the request.

Data acquisition is tableau-data's job and branding is tableau-brand's.

> The full STATE.md schema and the ordering / staleness / versioning rules live in > CONTRACT.md at the repo root. This skill restates only its own slice; intake.py > is the executable mirror of the contract it enforces.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: laviDrori0702
• Source: laviDrori0702/tableau-dashboard-creator-skill
• License: MIT

Install and usage instructions live in the source repository linked above.