| name | review-slide-flow |
| description | Review a Marp presentation’s logical flow from the audience’s point of view. Use when checking section order, slide-to-slide reasoning, unanswered questions, hidden prerequisites, promise fulfillment, transitions, or whether a proposed reordering will make a talk easier to follow. |
Review Slide Flow
Trace the reasoning an audience must reconstruct in real time. Find the smallest changes that restore continuity without adding decorative transition slides.
Inputs
Read the target Markdown completely. If present, also read vendor/3shake-marp-templates/.claude/rules/slide-writing.md. Infer the intended audience, duration, and central question from the deck and nearby documentation. State uncertain assumptions; do not stop unless different assumptions would reverse the recommendation.
Number slides by rendered order and record each title. Refer to evidence as slide N, “Title” plus a short quotation or paraphrase.
Workflow
- Write the talk’s central question and proposed answer in one sentence each. If either cannot be stated, record that as a critical finding.
- Build a promise ledger from the title, agenda, opening questions, and explicit “what you will learn” statements. Locate where each promise is answered or mark it open.
- Partition the deck into sections by conceptual purpose, not visual separators. Give every section one job.
- For each section boundary, identify the reasoning relation: consequence, contrast, decomposition, escalation, example, objection, synthesis, or application.
- For each adjacent slide pair, complete: “Because the audience just learned ___, they are now ready to consider ___.” A weak or impossible completion exposes a gap.
- Check prerequisites: terms, evidence, actors, constraints, and prior conclusions must appear before they are used.
- Model the audience’s likely question after each major claim. Verify that the next slide answers it or deliberately postpones it with a clear reason.
- Test the ending against the promise ledger. A summary must resolve the opening question without introducing a new thesis.