2654400439/Paper2Seminar / skills/paper-ppt-orchestrator
paper-ppt-orchestrator
VerifiedBuild complete fixed-layout academic-paper presentations from a PDF with bundled PPTX and one-page HTML templates. Use for full paper reading, complexity-aware slide budgeting, subsection coverage mapping, figure extraction, TikZ or Matplotlib evidence visuals, deterministic PPTX assembly, and final QA. Trigger for requests to turn a paper PDF into a discussion-group deck, especially 12-16+ page systems, security, networking, or measurement papers.
references/visual-selection.md
View on GitHub# Visual selection
## Decision order
Choose the visual after the slide purpose and evidence are fixed.
1. Does the paper contain the exact evidence or method image the audience must inspect? Use `paper_asset`.
2. Does the slide explain a synthesized process, mechanism, relationship, or comparison not available as a readable paper figure? Use `tikz`.
3. Does the paper provide at least two comparable exact values in text or a selectable table, and does their relationship matter? Use `data_redraw`.
4. Is the slide conceptual background where a realistic visual improves orientation without carrying factual detail? Use `imagegen` when capability preflight marks it available. Otherwise, if licensed web-image search is available, use `external_image`.
5. Is the slide a reflection, limitation, discussion, quote, or question that is clearer without a main image? Use `none`.
For a complete deck, satisfy a capability-aware diversity floor: one `tikz` visual, one faithful `data_redraw` visual, and at least one conceptual visual only when preflight selects `imagegen` or `external_image`. When preflight selects `tikz_or_none`, preserve that capability waiver and do not block delivery or invent an unavailable asset. Meet the applicable requirement during visual routing by choosing semantically appropriate slides. Do not force a mode onto an unsuitable slide or weaken evidence fidelity. P0 records this as a planning constraint and does not add a separate counting-only gate.
## `paper_asset`
Use for:
- Original method overview diagrams.
- Experimental charts and tables.
- Qualitative examples that must remain faithful to the paper.
- Results where values, axes, or legends are evidence.
Requirements:
- Asset must exist in the selected extractor's manifest/metadata or have an explicit manual crop record.
- Review the crop at the final slide size.
- Preserve or restate the source caption.
- Use `contain`; never distort.
- Record paper page and figure/table label.
- If the full table is unreadable, prefer a focused crop plus a textual takeaway. Do not silently redraw numbers without a verified data-redraw workflow.
## `tikz`
Use for:
- Multi-stage pipelines.
- Architecture summaries.
- Causal chains and threat models.
- Before/after or mechanism comparisons.
- Presenter-created synthesis across multiple paper sections.
Requirements:
- Read `tikz-layout.md` before generating.
- Declare the diagram as presenter synthesis when it is not copied from the paper.
- Keep `.tex`, `.pdf`, `.png`, and compile log.
- Compile with XeLaTeX and inspect the PNG.
- Prefer 6-10 nodes at most; move detailed explanation to slide text or notes.
Do not use TikZ to fabricate an experimental chart from visual estimation.
## `imagegen`
Use for:
- Conceptual background.
- A non-literal setting or metaphor.
- A cover or section opener when the template and user permit it.
Requirements:
- Never use on `results` slides.
- Avoid method slides unless the image is clearly decorative and cannot be mistaken for system structure.
- Prompt for no text, labels, logos, watermark, or numerical content.
- Save the prompt, chosen output, generation date/capability, and hash when practical.
- Inspect for misleading details, malformed objects, accidental text, and brand conflicts.
- Add alt text based on what the final image actually contains, not only the prompt.
If the execution tool is absent or returns a definitive unavailable/authentication error, record `imagegen: unavailable` and stop retrying that route. Use the capability report's next route.
## `external_image`
Use only as the conceptual-image fallback selected by capability preflight. It is not a substitute for a paper figure, experimental chart, or precise architecture diagram.
Requirements:
- Never use on `results` slides. Avoid method slides unless clearly decorative.
- Search for the intended concept, then open the source page and verify explicit reuse terms. Do not use a search-results thumbnail as the asset or source record.
- Prefer public-domain, Creative Commons, or clearly licensed stock sources. Do not assume that an image being publicly reachable grants reuse rights.
- Record `search_query` and an `external_source` object containing `page_url`, `license_name`, `license_url`, and `retrieved_at`. Record provider, creator, direct asset URL, SHA-256, and transformations when available.
- Download the original or a suitably high-resolution derivative into `assets/external/`; crop or resize without changing its factual meaning.
- Inspect for watermarks, visible personal data, misleading brands, accidental text, and details that could be mistaken for paper evidence.
- Credit the source in notes, a small caption, or the QA/source manifest when the license or context requires it.
## `none`
Use for:
- Limitations and open questions.
- Presenter reflection.
- A short quote, code fragment, or decision question.
- A transition where an image would add no information.
Requirements:
- Provide a rationale in the deck plan.
- Keep `none` below 20% of content slides unless the user explicitly requests a discussion-heavy deck.
- Do not leave an empty image placeholder on the final slide.
## `data_redraw`
Use only for a small set of exact values already available while the agent reads the paper and only when the chart communicates a relationship.
- Write a `data-viz.json` conforming to `data-viz.schema.json`.
- Require at least two comparable data points. Prefer a cross-method comparison, cross-dataset comparison, trend, distribution, heatmap, or compact result table.
- Keep one isolated value such as "up to 219x" in the slide text; a one-bar chart adds no information.
- Use paper text or selectable PDF table text. Do not use OCR.
- Do not estimate values from line height, bar length, or unlabeled geometry.
- Render with `render_data_viz.py` to a 2.1:1-2.7:1 PNG.
- If the spec exceeds 8 categories, 3 series, 6 table rows, or 5 table columns, simplify or use the paper asset.
- Prefer the original paper figure when it is already readable and carries axes, confidence intervals, or experimental context that a redraw would discard.
- Keep the spec and generated manifest beside the PNG.
- At least one `data_redraw` in a complete deck must be a faithful redraw of exact selectable paper values, with the original comparison semantics preserved. It may simplify presentation, but it must not estimate, interpolate, or alter the source values.
## P0 layout
All modes use `content_stacked`. Generate or crop visuals for the fixed lower image region rather than changing slide layout.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |