← ClaudeAtlas

report-designlisted

Runs PROFILING and DESIGN for the REPORT tier. Explores the data, reads it through the business, and turns that into a layout agreement of pages and questions before any page gets built. Covers both a new report and a change to one that already exists. Use after the brief is agreed and before building a report.
DiLoretoT/faw · ★ 0 · Web & Frontend · score 68
Install: claude install-skill DiLoretoT/faw
# Report design Between the brief and the first page there are two phases, and they exist because a report built without them is precise about questions nobody asked. The brief already settled what the report is for, who opens it and which questions it has to answer. What is still missing is whether the data can answer them, and what shape the answer takes. ## The three hats, in order, on the same context This is one process in three passes, not three separate reviews. Each pass reads what the previous one found and builds on it. Restarting from scratch on each one loses what was learned and asks the user again for what they already said. **Do not delegate these passes to separate agents.** The value is in the accumulation, and a subagent starts without the context that makes the next pass worth anything. The one delegation FAW does make is validation, precisely because there breaking the context is the point. ### First pass: what is actually in the data This is PROFILING, and it is read-only. Measure the model the report is going to consume: - Tables and their grain. What one row means in each. - Row counts, and the business totals the report will have to reproduce. - Measures that already exist. A report that needs a measure nobody wrote is a MODEL tier ticket, not something to solve inside a visual. - Cardinality of every dimension used for slicing. A slicer over a high-cardinality column is a decision, not a default. - Data quality where it affects an answer: