TEST-RESULTS: reel-scriptwriter
The full quick-check results of the reel-scriptwriter Skill: how it was run, what passed and what failed.
Date: 2026-10-14 · Kind: smoke test, not proof. One run (n=1), one input. No defect rate is claimed.
How: the agent that wrote the skill followed its SKILL.md step by step in the same session, on a fictional small business (a home bakery, «مخبز سنبلة»). It was not run in a fresh Claude session, so automatic triggering was not tested, and the author checking their own skill is a weaker test than an outside run.
| Run | Input | Result |
|---|---|---|
| 1 | examples/input.md: 45 s Reel, «4 things before ordering a cake», Iraqi voice with 2 sample sentences, 2 facts (5-day notice policy, a cake ordered 7 days ahead), a free PDF that exists | examples/expected-output.md: 6-beat script table, facts table, matching caption, self-review 8.5 → 9.5 in two passes |
What worked: the skill's «2 points in 30-45 s» rule caught a conflict with the idea (4 points): the script kept 2, turned 1 into the warning, and moved the 4th into the PDF, and said so. Only the 2 facts from the input were used. The self-review caught that the first draft was 68 words (about 30 s, not 45) and the fix added lines without new numbers. Keyword «قائمة» is the same in voice, screen and caption.
What failed or is weak: the length is still an estimate (not read aloud, not timed). Words per second (about 2-2.5) is a planning estimate, not a measured rule; the skill says so. The dialect match rests on 2 sentences.
Not tested: a formal-Arabic voice; 60 s with 3 points; an input with a missing source (should come out «مسودة»); a fresh Claude session or Claude.ai.