What problem does it solve? Session recordings are persuasive but easy to misread: teams watch a few replays, generalize from anecdotes, and ship findings nobody can verify. This Skill turns a pile of recordings into a short, graded findings table where every claim carries evidence coordinates a colleague can check in two minutes. ## Core Features & Use Cases - Question-scoped triage: refuses to open a player until a decision-relevant question exists, then classifies recordings from the index so only 5-10% of the corpus gets watched. - Four-grade evidence ladder: every claim is labeled MEASURED, RECONSTRUCTED, OBSERVED, or INFERRED, with agents forbidden from awarding themselves OBSERVED. - Dead-click and rage-click gate: classifies copy-selects, reading taps, false affordances, and slow controls before citing any count as a defect. - Use Case: Given a PostHog or local replay export showing drop-off at a report page, triage 159 recordings down to a dozen candidates, fan out subagents to extract event timelines, watch the survivors, and publish a 3-5 row findings table with grades, timestamps, and owners. ## Quick Start Ask the assistant to review the session recordings in this folder and produce a graded findings table explaining why users drop off before generating a report.
