agentsclimarketplace

Rebuild image as pptx

Skill chai798228814/rebuild-image-as-pptx/skills/rebuild-image-as-pptx

High-fidelity reconstruction of reference images, infographics, dashboards, one-page reports, posters, and slide screenshots as editable PowerPoint files. Use when Codex is asked to convert images into PPT/PPTX, reproduce a design one-to-one, rebuild editable text, tables, charts, cards, or diagrams, or update an editable PPT from a revised reference. Also use for Chinese requests such as “图片转可编辑PPT”, “截图还原PPT”, “一比一复刻”, “看板转PPT”, or “海报转PPT”. Preserve the source ratio, content, hierarchy, palette, density, and layout; render, audit, and perform text-consistency and typo review before delivery.From its SKILL.md

Install
npx -y skills add chai798228814/rebuild-image-as-pptx --skill rebuild-image-as-pptx

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 28 days oldThe repository was created 28 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

8.4 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

Rebuild Image as Editable PPTX

Reconstruct the reference as PowerPoint objects. Do not use the reference image as a full-slide background or claim completion without inspecting the exported PPTX. When general presentation-design defaults conflict with the supplied reference, faithful reconstruction of the reference takes precedence unless a technical or safety constraint makes it impossible.

Non-negotiable build contract

Apply every requirement below to each requested slide and to the deck as a whole:

  1. Generate a complete PowerPoint deck with the built-in Presentations skill.
  2. Reproduce the supplied image design as closely as possible: layout, palette, title hierarchy, content placement, chart style, card style, background, spacing, and visual emphasis.
  3. Do not redesign the source, substitute a new theme, or change colors without explicit user approval.
  4. Base slide content on the supplied design images and, when one exists, the confirmed outline. When multiple images are supplied, reconstruct every confirmed page in the requested order.
  5. Keep all text editable. Never satisfy the task by pasting the complete page image into PowerPoint.
  6. Recreate charts, flows, architecture diagrams, cards, timelines, tables, connectors, and other structured visuals with native PowerPoint objects whenever practical.
  7. When a complex visual cannot be reproduced exactly, use the closest editable PowerPoint shapes, gradients, vector icons, lines, and connectors while preserving its role and composition.
  8. Deliver a downloadable, editable .pptx only after self-validation: render the saved deck, compare every slide with its reference, verify all wording and data, complete a typo review, and fix missing text, wrong colors, clipping, overlap, overflow, broken connectors, or font substitution.

Treat these requirements as acceptance criteria, not optional preferences.

Required tools and routing

  1. Use the built-in Presentations skill for every PPTX build or edit.
  2. Use @oai/artifact-tool from JavaScript ES modules; do not use python-pptx.
  3. Treat a supplied image as the only visual source unless the user explicitly requests redesign or extra assets.
  4. When updating an existing PPTX from a revised image, preserve the existing file identity when possible and change only fields proven different by the new reference.

Workflow

1. Inspect and inventory the reference

  • Read the image at original resolution and record its pixel width, height, and aspect ratio.
  • Inventory every visible object: headers, labels, cards, tables, charts, connectors, icons, footers, shadows, borders, and background effects.
  • Transcribe all text and numeric data before authoring. Treat OCR as a hint only; manually verify every Chinese character, number, percent sign, decimal, unit, pct, and 个百分点.
  • Save the confirmed transcription as a required-text list. For revisions, also save superseded wording as a forbidden-text list.
  • For a revised reference, diff it against the current PPTX or earlier reference. List every changed phrase and value, including small regional-name or denominator changes.
  • Ask only when a genuinely unreadable item would materially alter the deliverable. Otherwise continue with the closest faithful reconstruction and flag uncertainty.

2. Set the slide geometry

  • Match the source aspect ratio. Use the source pixel dimensions as the artifact-tool slide size when practical.
  • If the user gives an exact ratio, that ratio overrides the image metadata.
  • Preserve source margins and module proportions. Do not silently convert a 3:1 dashboard into 16:9.

3. Rebuild with editable objects

  • Make all text editable text boxes.
  • Build cards, banners, borders, dividers, bars, timelines, flows, and diagrams with native shapes.
  • Rebuild tables with native cells or aligned editable shapes. Rebuild simple charts with native chart objects or editable bars and labels.
  • Keep icons as separate vector/SVG objects when native shapes are impractical. Icons may be images; body text may not.
  • Never insert the complete reference image as the slide background or a full-slide picture.
  • Give important objects stable names for later inspection and targeted edits.
  • Preserve the source content, spelling, hierarchy, palette, emphasis colors, and alignment. Do not “improve” or redesign unless requested.

4. Handle Chinese typography

  • Use the font requested by the user. Otherwise preserve the apparent source family; default to Microsoft YaHei for modern Chinese business dashboards.
  • If the render environment lacks the requested CJK font, configure a local CJK fallback for QA while keeping the requested PowerPoint typeface in the PPTX.
  • Treat missing Chinese glyphs, dots, blank labels, or Latin-only fragments as build failures.

5. Export and validate

Render and inspect the exported PPTX, not only the in-memory preview.

  1. Render every slide with the Presentations skill's render_slides.py.
  2. Run slides_test.py and fix all unintended overflow.
  3. Create a visual comparison with scripts/compare_renders.py. Inspect the reference, render, 50% overlay, difference image, changed-pixel ratio, and changed bounding box at full size; do not treat one global score as sufficient.
  4. Run scripts/audit_pptx.py with the reference image, required-text list, forbidden-text list when applicable, --text-output, and --allow-missing-typo-review to extract the final PPTX text for review.
  5. Compare the extracted PPTX text with the confirmed transcription. Treat any PPTX/reference mismatch as a reconstruction error and fix it automatically.
  6. Review the extracted text for suspected Chinese typos, repeated or missing characters, inconsistent terminology, punctuation, units, dates, region names, and visually similar OCR substitutions. Save the result as typo-review.json using the schema in references/qa-checklist.md.
  7. If the PPTX matches a suspected typo in the reference, preserve the source by default and report the source wording, suggested correction, and slide number instead of silently changing it.
  8. Rerun scripts/audit_pptx.py with --typo-review-file typo-review.json. Verify all changed fields, verify superseded text is absent, and repeat until all mandatory checks pass.

If a legitimately image-dominant source triggers the raster-coverage safeguard, inspect it manually and use --allow-high-raster-coverage --raster-override-note "<reason>". Never use the override to permit a flattened reference-page copy.

Read references/qa-checklist.md before final delivery.

Acceptance rules

  • Aspect ratio matches the source or explicit user ratio.
  • All visible text is present, editable, and readable.
  • Every displayed number matches the reference exactly.
  • Text consistency and typo review is complete. Reconstruction errors are fixed; suspected source typos are explicitly reported.
  • No old wording remains after an update.
  • No full-slide reference image is embedded.
  • No unintended clipping, overlap, overflow, missing glyphs, broken connectors, or collapsed charts remain.
  • The exported PPTX has been rendered and visually compared at full size.

Do not deliver a PPTX that merely opens or passes an overflow test if its text, layout, or styling visibly differs from the reference.

Final delivery

  • Save a new file for a new reconstruction; replace the same persistent file for an explicit revision unless the user asks for a copy.
  • Return only the final PPTX, not QA renders, scripts, or intermediate files unless requested.
  • State the ratio, editability, validation result, and typo-review result concisely. If suspected source typos remain, list them as warnings with slide numbers and suggestions. Do not claim “one-to-one” unless the visual comparison supports it.

What ships with it: 4 files

22.3 KB alongside SKILL.md, 2 of them executable

agents/

references/

scripts/

Gives 0 of the 12 instructions most pdf office docs skills give in ~1.6k tokens

Counted across 636 of the 690 authors here whose files we hold, read 2026-08-07

  • Extract text or tables using pdfplumber or pdftotextin 89 of 636, across 23 files
  • Create new PDFs using reportlabin 83 of 636, across 16 files
  • Read forms.md before filling out PDF formsin 80 of 636, across 13 files
  • OCR scanned PDFs using pytesseract and pdf2imagein 77 of 636, across 10 files
  • Use qpdf to merge or split PDFs or large filesin 70 of 636, across 3 files
  • Use Excel formulas instead of hardcoded calculated values or Python calculationsin 68 of 636, across 13 files
  • Unpack, edit, and repack XML for existing documents or presentationsin 63 of 636, across 8 files
  • Document sources for all hardcoded valuesin 61 of 636, across 9 files
  • Write minimal, concise Python code without unnecessary commentsin 59 of 636, across 7 files
  • Run the recalculation script (recalc.py) after adding or modifying formulasin 59 of 636, across 7 files
  • Fix all identified formula errors and recalculate before finishingin 58 of 636, across 6 files
  • Format years as text stringsin 57 of 636, across 5 files

Said here and by no other author read

  • generate a complete PowerPoint deck
  • reproduce the supplied image design closely
  • base slide content on supplied design images
  • recreate visuals with native PowerPoint objects
  • match the source aspect ratio
  • transcribe all text and numeric data

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.