Gorden Image PPT Generator
Turn any topic into a luxurious, information-dense, image-per-slide deck, ready to be converted into a fully editable PPTX downstream.
No malicious behavior observed: zero-data-fabrication rules, mandatory imagegen-manifest. evidence trail, and pre-run user confirmation of style/audience/page count are positive safety design. Deducted for: no formal license metadata (only informal attribution terms in README), unverified publisher identity, incomplete disclosure of full data flow (reads/writes to $CODEX_HOME, external image-generation service), and no rollback/cleanup mechanism.
The A1–A5 workflow is self-consistent; compose_pptx.py gives clear errors (_die/stderr) for missing files, invalid hex, missing Pillow, and the manifest gate prevents silent failure. Deducted for: static review cannot reproduce execution, no committed tests covering key paths, and retry/failure feedback for the core imagegen dependency depends mostly on runtime behavior.
Clear scenario (topic → image-based deck), declared boundaries (image-only output, fail-stop if imagegen unavailable), and explicit trigger conditions in the description. Deducted for: README restricts usage to Codex; hard dependency on GPT image generation and vision creates reachability/fit concerns for mainland-China users; multi-runtime support is theoretical without dedicated adaptation.
Good layered documentation (SKILL.md → reference guides/runtime notes → scripts), output conventions, schema and examples. Deducted for: no version/changelog, missing license metadata (informal terms only), unclear maintenance ownership and update path, and hidden dependencies on the not-reviewed GordenImage2PPTX skill (e.g., green-screen constraints).
The core task (image-based slides with verbatim text composed into a .pptx) is sensibly designed with a detailed prompt system and QA rules; the A5 script directly produces usable image-deck pptx with clear marginal value over manual design. Deducted for: static review cannot verify image/text rendering quality, image-based output is inherently limited for editing, showcase screenshots are not independently verifiable, and cost/benefit (heavy image-generation quota) is unquantified.
Repository contains source, reference gallery, README showcase images, and a manifest mechanism that mandates generation evidence — auditable primary material. Deducted for: no third-party reproduction, no CI/test coverage of this skill's key paths, and marketing claims ('strongest PPT skill ever') not separated from verifiable conclusions.
- Hard dependency on Codex built-in imagegen and GPT image/vision capabilities; other runtimes are unadapted, and overseas service reachability may affect mainland-China users.
- Static review only: text-rendering accuracy and composed pptx usability are unverified; run a small trial first and check imagegen-manifest..
- License metadata is missing; only informal attribution terms exist in README — confirm terms before commercial use.
- Output is full-bleed bitmap slides; editable pptx requires the not-yet-reviewed GordenImage2PPTX skill, creating coupling risk.
- Image generation consumes substantial quota (README claims ~10% of Plus 5-hour quota per image conversion); assess cost before large-scale use.
What does this skill do, and when should you use it?
GordenImagePPTGen is the first of three skills in the GordenSuperPPTSkills collection. Given a topic or content, it designs a per-slide outline, writes self-contained image-generation prompts containing all verbatim text, calls GPT's image generation (Codex's built-in imagegen) to render each slide as one image, and composes the images into a full-bleed image-based .pptx via compose_pptx.py. The skill deliberately forbids drawing slides with code and requires a per-slide imagegen-manifest. as acceptance evidence. It only produces image-based decks; editable PPTX output requires the sibling skill GordenImage2PPTX. The author states it is adapted for Codex only; other runtimes with an image-generation API could work in theory but are unadapted.
It runs a five-stage pipeline: A1 confirms style, audience, slide count and language; A2 produces a structured outline. assigning each slide a distinct complex layout (Bento grid, funnel, layered architecture, etc.) with a unified color scheme; A3 writes a self-contained prompt per slide (all verbatim text, hex colors, layout); A4 calls the image-generation model per slide, copies outputs to the task directory and writes imagegen-manifest.; A5 runs scripts/compose_pptx.py against deck. to assemble each PNG into a full-bleed image-based .pptx. Defaults: 16:9, 11–20 slides, ~20+ information points per page, zero fabricated data, user's language for slides.
- Professionals who need a visually rich, high-density deck fast and accept one rendered image per slide
- Users with real data or existing content who require the AI to render strictly from source material without inventing figures
- Codex users planning to follow up with GordenImage2PPTX to get an editable PPTX
- Slide makers needing a non-Chinese language or a non-16:9 ratio such as 3:2
- Authors of knowledge-sharing or training material who want a different complex framework on every page (Möbius ring, concentric radar, fishbone, etc.)
What are this skill's strengths and limitations?
- Forces thick content before rendering: ~20+ information points per slide, a distinct complex framework per page, simple layouts explicitly rejected
- Zero data fabrication and verbatim text in prompts — output is a finished slide image, not an empty template
- Rigorous engineering constraints: per-slide imagegen-manifest. evidence, code-drawn slides and placeholders banned, hard stop on failure instead of a code-drawing fallback
- Supports any style (tech/business, cartoon, ink-wash, etc.), non-16:9 ratios, and self-contained prompts that allow reproducible single-page re-renders
- README explicitly says Codex only; other platforms are unadapted, and the skill hard-fails when imagegen is unavailable
- Output is an image-based PPTX — slide text is not editable without the downstream GordenImage2PPTX skill
- Generated images may contain wrong characters; SKILL.md only promises manual fixing after conversion to editable PPTX
- No open-source license file in the repo (only a credit-based commercial-use statement in the README); no test suite or third-party evaluation evidence
- Heavy use of green is discouraged to avoid conflicts with the downstream green-screen matting, adding a color-scheme constraint
How do you install this skill?
This is one skill inside a three-skill collection repo. In Codex, send the repo URL and ask it to install, or copy manually: cp -R GordenImagePPTGen "${CODEX_HOME:-$HOME/.codex}/skills/GordenImagePPTGen". Dependencies: pip3 install python-pptx pillow (the README's collection-level deps also include numpy). No license file is declared in the repo; the README says commercial use is allowed provided you credit the GitHub repo or the author @Gorden Sun.
How do you use this skill?
Codex only (README: requires GPT image generation and vision; suggested GPT 5.5 model with medium reasoning effort). Example prompt: "Use the GordenImagePPTGen skill to generate an N-page PPT on XXX, requiring a luxurious, high-density, complex layout." For an editable PPTX afterwards, use GordenImage2PPTX; for one-click end-to-end, use GordenSuperPPTSkill. Image-to-editable conversion is credit-heavy (README: roughly 10% of a Plus subscription's 5-hour quota per image), but that cost belongs to the downstream skill; this skill's own credit cost is not documented.