What problem does it solve? Writing PRDs often fails in two ways: AI assistants invent plausible-sounding content the user never said, and hypotheses get documented without any numerical evidence. This Skill enforces a faithful-documentation principle (only write what the user said, mark everything else as [TBD]) and then fills those gaps with real data pulled from email delivery tools, BI/SQL environments, and GitHub. ## Core Features & Use Cases - Faithful PRD skeleton drafting: Classifies user statements into explicit facts, implied directions, and unmentioned items, leaving [TBD] placeholders instead of guessing. - Multi-source data collection: Gathers evidence from email campaign history (send dates, open rates, CTR), cohort analysis via SQL, GitHub PRs/Issues, and internal documentation tools. - Campaign effect tables: Builds per-campaign weekly increment tables comparing overall trends against fixed cohorts, with warnings for same-week overlaps and Friday sends. - Claim-data consistency checks: Reports discrepancies between user claims and data back to the user instead of silently rewriting the PRD. - Use Case: A product manager wants to document the impact of three feature releases. The Skill drafts the PRD from their verbal explanation, generates SQL for cohort analysis, cross-references email send logs and GitHub release PRs, and produces a PRD where every number is traceable to a query or file. ## Quick Start Ask the assistant to draft a data-backed PRD from your verbal description of a feature launch, then have it fill the [TBD] sections using your email delivery exports and BI tool query results.
