eva-review

What problem does it solve? After publishing short videos, social posts, or articles, creators often cannot tell why a piece performed the way it did or what to test next. This Skill turns published-content data into one falsifiable hypothesis and one concrete next-post experiment instead of vague platform guesses. ## Core Features & Use Cases - Single-post review: Reads screenshots, metrics, and transcripts of one published piece, checks title/opening/body promise fulfillment, and outputs one hypothesis, one test variable, one metric, an observation window, and a falsification condition. - Batch pattern review: Groups comparable records by account, platform, format, and goal; with 10+ comparable records it surfaces up to three candidate patterns with counterexamples, and can produce a stage content-mix snapshot on request. - Result backfill: Matches new results to the original hypothesis and judges support / not support / inconclusive without overwriting prior records. - Persistent record library: After user authorization, maintains a per-account local archive of review records, backfills, and pattern reports under ./eva-review/. - Use Case: A creator shares backend screenshots of a published Douyin video; the Skill identifies the most likely bottleneck, proposes changing only the opening hook in the next post, and defines what result would disprove the hypothesis. ## Quick Start Say: help me review this published video with these backend screenshots and tell me what to test in the next post.