What problem does it solve? Product discovery often starts from fuzzy prompts or raw client transcripts with no structured way to turn them into validated jobs-to-be-done, architecture decisions, and research plans. This Skill runs a disciplined, human-in-the-loop discovery phase that converts raw input into durable product artifacts without inventing facts. ## Core Features & Use Cases - Ten-movement discovery workflow: Sequences challenge, run brief, research ingestion, straw-man JTBD drafting, dev review subagent, architecture drafting, ledger-driven grill, consolidated read-back, independence handoff, and a final write step. - Evidence discipline: Stages every source as an input batch with coverage rows, requires line citations for domain claims, tracks conflicts as ledger items, and binds decision records to falsifiable assumption records. - Human-gated decisions: Escalates blocking questions one at a time with recommended answers, never guesses demand-side answers, and serializes unanswered questions under an Awaiting answers section when no human responds. - Use Case: A product manager pastes a client interview transcript and asks the agent to run discovery; the Skill digests the transcript, drafts jobs and decision candidates, grills the human on blocking unknowns, and writes context.md, jtbd/ files, ADRs, decision records, ateam-plan.md, and research-plan.md. ## Quick Start Run the ateam-discovery skill on this client transcript to seed docs/product/ with jobs, decisions, and plans before any feature exists.
