GordenSuperPPTSkill — End-to-End PPT Pipeline
One pipeline that first generates rich image-based slides, then reconstructs them into a fully editable .pptx, delivering both outputs at once.
No destructive defaults, credential access, or covert exfiltration observed; manifests (generated_source/copied_to) give partial data-flow transparency; the skill stops when imagegen is unavailable instead of silently degrading. Deducted for: hard dependency on Codex/OpenAI image services with no privacy disclosure about uploading user content, and unclear license metadata (README only has custom attribution terms), leaving attribution and redistribution terms incomplete.
SKILL.md and pipeline.md are internally consistent with stage gates (missing imagegen-manifest fails stage 1) and explicit blocking behavior. Deducted for: static-only review; referenced scripts (compose_pptx.py, chroma_key.py) not present in reviewed evidence, no test suite, and OCR/chroma-key quality is an unverified assumption with uncontrolled failure modes.
Trigger table is clear (named sub-skill vs default Super), scenario boundaries described, good Chinese support. Deducted for: hard Codex + GPT imagegen requirement with no adaptation to other runtimes, full dependence on overseas services with questionable mainland-China reachability, and high quota cost (~10% of Plus 5h per image) not disclosed inside the skill docs.
Well-layered docs (SKILL.md + references + self-contained sub-skills), clear install notes, dependency list and directory structure. Deducted for: unknown license metadata (custom attribution terms only), no versioning or changelog, no FAQ/known-limitations section, maintenance ownership resting on a single author, and exaggerated marketing description ('史上最最最强').
The claimed end-to-end flow has a clear deliverables checklist and a four-layer reconstruction design; README includes example screenshots, and if it works as described the marginal value is significant. Deducted for: static review cannot verify actual editable-PPTX reconstruction quality; examples are not independently reproducible; 'fully editable' completeness is unverified and may require substantial manual correction.
Some auditable primary material exists (example images, deliverables lists, manifest requirements) placing it at the 'limited reproduction' tier. Deducted for: no CI workflows or committed tests covering key paths, claims resting mainly on author self-report and marketing language, and no cross-source corroboration or fact/inference separation.
- The skill is Codex-only and hard-depends on OpenAI imagegen; mainland-China network reachability is uncertain, and user content is uploaded to the image service, so users with sensitive content must assess privacy risk themselves.
- Cost is high: the author states each image consumes ~10% of a Plus subscription's 5-hour quota, making a full deck expensive.
- License metadata is unknown; README carries only custom attribution terms (commercial use must credit the source / @Gorden Sun). Confirm scope with the author before commercial use.
- This is a static review with no execution; the 'fully editable PPTX' reconstruction quality, OCR accuracy and chroma-key fidelity are independently unverified and outputs may need manual correction.
- No versioning, changelog or test evidence; the description is marketing-exaggerated, so claims should be validated by trial.
What does this skill do, and when should you use it?
This is the orchestration skill in the GordenSuperPPTSkills collection, chaining two sibling skills: image-PPT generation and image-to-editable-PPTX conversion. Stage 1 calls Codex's built-in imagegen model to render each slide as a finished image; Stage 2 extracts background, frame, icons, and text from each image and reassembles them into an editable pptx by coordinates. The skill itself implements nothing — it sequences the sub-skills, enforces quality gates, and bundles the deliverables. The README states the collection is Codex-only, since it depends on GPT image generation and vision.
Reads and fully executes the sibling skills GordenImagePPTGen (steps A1–A5) and GordenImage2PPTX (steps B1–B9) from the same directory. Stage 1 mandates a real imagegen call for every slide (PIL/SVG/native-shape fallbacks are explicitly banned), requires all real text verbatim in each image, and produces outline., prompts/, slides/NN-*.png, imagegen-manifest., and an image-based pptx. Stage 2 enforces a four-layer reconstruction per image — background, green-screen-keyed frame and icon layers (all must be imagegen-extracted), plus OCR text — recording per-page asset manifests, then composes an editable .pptx. Delivery is a single bundle of all intermediates plus both pptx versions.
- A user provides only a topic or content with no specific feature requested and wants a finished deck that both looks good and is editable
- Scenarios needing both an image-based deck (for preview/sharing) and an editable PPTX (for later edits)
- Marketing or teaching slides that demand high information density and complex layouts, produced quickly
- Codex users with the full skill set installed who want to skip manually invoking the two stages in sequence
What are this skill's strengths and limitations?
- Delivers image-based PPT, editable pptx, and all intermediates in one run, removing manual two-step work
- Strict built-in gates (imagegen-manifest. checks, mandatory four layers, no code-drawn fallback) prevent degradation to placeholder images
- Consistent aspect ratio and palette throughout, no repeated per-slide frames, well-structured reusable outputs
- Each skill directory is self-contained with scripts and references, easy to tune independently
- Hard dependency on Codex's built-in imagegen; on non-Codex runtimes the skill must halt, and no other platform is adapted
- Expensive: the README says converting one image costs about 10% of a Plus subscription's 5-hour quota
- No automated test suite; reconstruction quality depends on model image generation and vision; no license field
- Sibling skills must live in the same skills directory, so partial installs break the pipeline
How do you install this skill?
It depends on the other two skills in the repo, which must be installed into the same skills directory. Simplest: copy the whole repo over:
cp -R GordenImagePPTGen "${CODEX_HOME:-$HOME/.codex}/skills/GordenImagePPTGen"
cp -R GordenImage2PPTX "${CODEX_HOME:-$HOME/.codex}/skills/GordenImage2PPTX"
cp -R GordenSuperPPTSkill "${CODEX_HOME:-$HOME/.codex}/skills/GordenSuperPPTSkill"
Then install dependencies: pip3 install python-pptx pillow numpy. The README does not document Windows path differences.
How do you use this skill?
Codex only (README recommends the GPT 5.5 model at medium reasoning effort). Send the GitHub repo URL to Codex to install, then prompt for the full pipeline, e.g.: "Use GordenSuperPPTSkill to generate an N-page PPT about XXX — luxurious, information-dense, complex layout — and convert it to an editable PPT." If you explicitly want only image slides, or only image-to-editable conversion, invoke the corresponding sub-skill instead of this one.
How does this skill compare with similar options?
The source names no specific competitors, only self-describing as the "terminator of the AI PPT race." Compared with the common approach of drawing slides programmatically with python-pptx, its differentiator is insisting on per-page model-generated images reconstructed backwards — but no third-party comparison is cited in the source.