--- name: review description: Review a MathCoffee video project's mathematics, narrative, layout, timing and encoded delivery evidence without treating automated checks as human acceptance. --- Resolve `../../bin/mathcoffee` from this skill directory to an absolute quoted path. Run `status --workspace --json` and the appropriate `verify` command after inspecting `--help`. Read the actual brief, canonical authoring, resolved style and build/review records before concluding anything. For mathematics, check definitions, assumptions, calculations, counterexample boundaries and whether each geometric action comes from the same recorded state. For narrative, follow the real playback order: what question is open, what has been established, what obstacle remains, and why the next concept helps. Review a continuous transition rather than unrelated best-looking stills. For visuals, inspect actual encoded intermediate frames, longest labels, arrows, overlaps, font resolution, safe areas, semantic colors and formula holds. Apply the selected preset, not a style from an unrelated historical episode. Portrait defaults omit burned narration captions while retaining narration and necessary mathematical text. For sound, check actual sample-based timing, intact sentence ends, neighboring voice joins, music ducking and the encoded audio track. Polyphones, English letter names, subscripts and mixed language need real listening. Forced alignment, ASR or embedding scores don't certify pronunciation or naturalness. Check all audience text through the final black frame. Academic references may be audience-facing; engine names, paths, internal process, hashes and acceptance claims belong in project reports. Compare current source identities with the rendered artifacts to catch stale captions or closing audio. Return actionable findings with scene/sentence IDs, timestamps and evidence, plus the exact scope reviewed. Preserve separate statuses for automated technical validation, independent mathematics/narrative review, visual sampling, actual human listening and audience feedback. Leave unavailable evidence pending. A review request alone doesn't authorize rewriting the source, installing dependencies, or publishing.