Update to AI Board Lite

{
  "description": "Reviews a presentation the way a board member reads it. Checks whether the titles alone tell the story, whether every number is sourced and consistent across slides, and what a critical reader will ask. Returns a verdict, slide-by-slide fixes and the questions to prepare for. Use for \"review this deck\", \"is this ready for the board\", \"check mijn presentatie\", \"wat ontbreekt er aan dit deck\".",
  "included_files": [],
  "name": "review-deck",
  "skill_md_contents": "---\nname: review-deck\ndescription: Reviews a presentation the way a board member reads it. Checks whether the titles alone tell the story, whether every number is sourced and consistent across slides, and what a critical reader will ask. Returns a verdict, slide-by-slide fixes and the questions to prepare for. Use for \"review this deck\", \"is this ready for the board\", \"check mijn presentatie\", \"wat ontbreekt er aan dit deck\".\n---\n\n# Review Deck\n\nA deck fails in the room for three reasons: the story only works if someone\ntalks over it, a number on slide 3 does not match slide 9, or the obvious\nquestion has no answer. This review finds those three before the meeting does.\n\n## Ground rules\n\n- Review only the deck the user shares, and quote slide numbers for every\n  finding.\n- Quote numbers exactly as they appear. Never correct a number yourself: point\n  at the conflict and ask which one is right.\n- Be direct. A review that hedges every point is useless the night before a\n  board meeting.\n- Answer in the language the user writes in.\n  The example below is in English; translate its headings too.\n\n## When to use\n\nThe user shares a `.pptx` or `.pdf` and asks \"review this deck\", \"is this ready\nfor the board\", \"what will they ask\", \"check mijn presentatie\", \"wat ontbreekt\ner aan dit deck\".\n\n## Steps\n\n1. **Ask who the audience is** and what the deck should achieve, if that is\n   not clear. A deck for approval is judged differently from a deck for\n   information.\n2. **Extract the text per slide** with whatever this environment offers\n   (`markitdown`, `python-pptx`, a PDF reader). If you cannot read the file,\n   ask for a PDF export or the slide text. Never review a deck you could not\n   read.\n3. **Read the titles only, in order.** Write down the story they tell. If it\n   has gaps or the main point only appears on slide 7, that is finding one.\n4. **Check every number:** is it sourced, and does the same metric have the\n   same value everywhere it appears? List every conflict with both slide\n   numbers.\n5. **Check every slide for one message.** Flag slides that make two points,\n   slides with a label instead of a title, and slides that prove nothing.\n6. **Write the five questions** a critical board member will ask, each with\n   the slide it comes from and whether the deck currently answers it.\n7. **Give the verdict first**, then the fixes in order of importance.\n\n## Output format\n\n```markdown\n## Verdict\nNot ready yet. The recommendation is sound but sits on slide 8, and revenue\nappears with two different values.\n\n## Fix first\n1. **Move the recommendation to slide 1.** Right now the board reads seven\n   slides before learning what you want.\n2. **Revenue conflict:** slide 3 says €4.1M, slide 9 says €3.9M. Which one is\n   right, and is one of them a forecast?\n\n## Slide by slide\n| Slide | Title today | Problem | Fix |\n|---|---|---|---|\n| 2 | \"Market\" | Label, not a message | \"Our segment grew; our share did not\" |\n| 5 | (two charts) | Two points on one slide | Split into 5a and 5b |\n\n## Questions to prepare for\n1. What happens if the pilot misses its date? *(slide 6, not answered)*\n2. Why Service and not Sales? *(slide 4, answered)*\n\n## Limits of this review\nText only. Charts and images were not checked.\n```\n\n## Limits\n\n- Text only. A chart that contradicts its own title is not caught. Say so in\n  every review.\n- This skill reviews. It does not rewrite or rebuild the deck unless the user\n  asks for that afterwards.\n"
}