What does this skill do, and when should you use it?
This Codex skill addresses a common pain point: you have slide screenshots but want a truly editable .pptx, not full-slide images pasted as backgrounds. It decomposes each screenshot into three layers — AI-generated text-free PNG assets for complex visuals, native PPT shapes for simple geometry, and native text boxes for all readable text. The workflow runs in three phases (pixel-level analysis, visual asset generation, PPT assembly and validation), with per-slide render-and-compare loops to keep output faithful to the source's layout and palette. The skill is a pure instruction document (SKILL.md) that relies on the host platform providing image generation and a PptxGenJS presentation skill.
Reads user-supplied slide screenshots; catalogs every visible element per slide (type, percent-based bounding box, layer, implementation) with a completeness self-check to catch small icons and in-card illustrations; calls $imagegen to generate text-free PNG assets for Layer A elements (named s{slide}_{role}.png); builds the deck with a PptxGenJS-based skill ($slides or presentation-skill) in strict z-order, converting percent coordinates to a 13.333×7.5-inch canvas and extracting exact colors and fonts from the source; finally runs structural checks (multiple text frames and shapes per slide, no picture covering over 85% of the slide), render comparison, and an editability check. Outputs include _analysis.md, an assets/ folder, _phase2_assets.md, validation_report.md, and the final .pptx.
- Professionals who lost the original deck and only have exported screenshots but need an editable version back
- Design or sales teams who received someone else's slide images and must adapt the content
- Converting AI-generated slide mockups into real PowerPoint files
- Training ops with many instructor course screenshots that need batch reconstruction into editable courseware
- Consultants who need rebuilt slides that preserve the original look while remaining easy to revise
How do you install this skill?
- SKILL.md content is truncated at the end; verify the complete file after installation.
- Core function depends on Codex-specific $imagegen and @presentations skills; not directly portable to other platforms, and Chinese users should confirm service reachability in their network environment.
- Using $imagegen sends slide image content to an external image-generation service; do not use with confidential or sensitive material.
- Per-slide rendering and human comparison are required; actual time cost may exceed expectations and nothing has been execution-verified.
- Publisher is unverified, with no versioning or changelog; track the repository yourself for updates.
- Shell / CLI
- Network access
- Local filesystem
image generation tool ($imagegen)PptxGenJS-based presentation skill ($slides / @presentations)
Copy the skill folder into your Codex skills directory: git clone https://github.com/w1163222589-coder/slide-image-to-editable-pptx.git, then cp -r slide-image-to-editable-pptx ~/.codex/skills/. Alternatively install via CC Switch: Skills panel → Add from GitHub → paste the repo URL. The source does not document an official Claude Code installation path.
git clone --depth 1 https://github.com/w1163222589-coder/slide-image-to-editable-pptx.git ~/.claude/skills/slide-image-to-editable-pptxGenerated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Put your slide screenshots in source_slides/. The recommended three-prompt workflow: Prompt 1 runs Phase 1 (including the Step 1.4 completeness self-check) and outputs the full element inventory for review; Prompt 2 runs Phase 2 to generate assets and outputs the asset report — do not proceed until confirmed; Prompt 3 runs Phase 3, using @presentations to rebuild each slide per _analysis.md, starting with slides 1-5 and rendered screenshots for review, then continuing in batches of 5. Image generation capability and a PptxGenJS presentation skill must be available during the run.
What are this skill's strengths and limitations?
- Clear three-layer principle: never bake text into images, never approximate complex visuals with crude shapes, never use full-slide screenshots as backgrounds
- Forces per-slide analysis with a completeness self-check targeting missed small icons and in-card illustrations
- Mandates pixel-level position measurement and exact color extraction, preventing 'new template' drift
- Per-slide render comparison catches layout deviations before moving on
- Four documented failure modes (new template, generic motif spam, crude shapes, baked text) each with symptom, cause and fix
- Hard dependency on a platform-provided $imagegen tool and a PptxGenJS-based presentation skill; the pipeline breaks without them
- By design, complex image assets are only replaceable (move/resize/delete), not internally editable
- Font handling gives approximations only (Microsoft YaHei, Cambria Math); exact source fonts may not be reproduced
- No automated test suite in the repo; compatibility claims beyond the author's table (e.g. Claude Code) are not third-party verified
- Requires human review between the three phases — not a one-click fully automated flow
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Slide Image to Editable PPTX this page | 46 · Use with care | ★ 196 | 4mo ago | MIT |
| GordenImage2PPTX — Slide Image to Editable PPTX | 48 · Use with care | ★ 2.1k | 4mo ago | — |
| Image to Editable PPT | 54 · Use with care | ★ 2.9k | 1mo ago | MIT |
| PPTX Presentation Skill ✓ Anthropic · Official | 54 · Use with care | ★ 181k | 16d ago | — |
| PPTX Presentation Skill (Open Cowork) | 69 · Recommended | ★ 2.2k | 12d ago | MIT |
The README contrasts three common alternatives: pasting screenshots as backgrounds (zero editability), asking AI to 'recreate' the PPT (right topic, wrong layout — a new template, not a reconstruction), and manual rebuilding (hours per slide). This skill positions itself as combining visual fidelity with editability.
How did FollowSkills review this skill?
Prompt-only skill: no scripts, no permissions, no destructive effects; single data flow (source image → generated PPTX). Deducted: user slide images are sent to the external $imagegen service with no data-flow/privacy disclosure, confirmation, or rollback notes; publisher identity unverified.
Three-phase workflow is self-consistent with self-checks and validation, but SKILL.md is truncated mid-sentence at the end ('no t'), the skill fully depends on host-provided $imagegen/@presentations skills, and there are no scripts, tests, abnormal-input handling, or defined failure feedback; static review cannot execute anything.
Scenario and trigger description are clear, with declared non-fit boundaries (no full-slide screenshot backgrounds); deducted for hard dependency on Codex-specific $imagegen/@presentations, uncertain portability, and no evidence of environment reachability for Chinese users who need equivalent image-gen and PPTX skills.
Well-layered bilingual README with failure-mode tables, install instructions, and a clear MIT license; deducted for missing versioning/changelog, unclear maintenance ownership, YOUR_USERNAME placeholder in install commands, truncated SKILL.md text, and no FAQ.
Real pain point addressed and the three-layer decomposition methodology is sound, claiming directly usable editable PPTX output; deducted because no executable representative output is verifiable, quality depends heavily on image-generation results and per-batch human review, and cost/benefit is unproven.
Only author-supplied screenshots (content not verifiable statically); no test suite, no third-party reproduction, no CI evidence covering the skill's key paths; claims and inferences are not separated.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →