← Back to Skills Marketplace
javastarboy

ppt-produce-bycodex

by AGICaptain · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
48
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install ppt-produce-bycodex
Description
Create and iterate enterprise application software solution PPTs as full-page PNG slides and optional PPTX decks. Use when the user wants to start, design, g...
README (SKILL.md)

PPT Produce by Codex

Use this skill to run a disciplined enterprise-solution PPT production workflow: clarify project context, create a storyline and page plan, generate full-page PNG slides with imagegen/gpt-image-2, review quality, and assemble PPTX only when useful.

For the complete reusable playbook, read references/ppt-production-guide.md when planning or executing a real deck. Keep this SKILL.md as the quick operating procedure.

Required Intake

Before creating files or generating slides, confirm these fields. If any are missing and cannot be inferred from the current workspace, ask the user for them first.

  • Project name
  • Project root directory
  • Solution topic or PPT title
  • Target audience or customer type
  • Current output version, such as v1, v2, or v3
  • Desired deliverables: PNG only, PPTX, outline MD, scheme MD, contact sheet, demo prototype, or a subset
  • Available source materials: meeting notes, docs, screenshots, old PPT, reference images, brand assets, existing outline

If the user gives a project root, keep all new artifacts under \x3Cproject-root>/workspace/. Do not create root-level output/, outputs/, tmp/, or scripts/ folders.

Default Workspace Layout

Use this layout unless the project already has a compatible convention:

  • workspace/deliverables/ for final outputs
  • workspace/assets/ for reusable source assets and copied references
  • workspace/runs/ for generation batches, staging outputs, backups, and contact sheets
  • workspace/tmp/ for temporary prompts, OCR, unpacked PPTX checks, and scratch scripts
  • workspace/scripts/ for reusable generation and verification scripts

Report full absolute paths for every created or modified file.

Workflow

  1. Inspect current files and existing versions. If a higher version directory exists than the user mentioned, ask which version is authoritative.
  2. Collect missing intake fields before generating.
  3. Read references/ppt-production-guide.md for detailed guardrails.
  4. Build or update the PPT storyline: background, positioning, overall architecture, core capabilities, scenarios, value, implementation path.
  5. Create a page plan with page title, page type, visible text, source assets, and generation method.
  6. Generate 2-4 sample pages before batching when a visual direction is not yet approved.
  7. Use small batches, staging folders, and contact sheets. Do not overwrite approved PNGs without a backup.
  8. Only assemble PPTX when the user asks or after a meaningful batch is confirmed; otherwise ask whether to generate PPTX.
  9. If business content changes, sync the scheme MD / PPT outline / PNG page content so future delivery and demos stay consistent.

Image Generation Rules

When generating or redrawing PPT PNG pages:

  • Use the imagegen skill at /Users/javastarboy/.codex/skills/.system/imagegen/SKILL.md.
  • Use gpt-image-2 unless the user specifies another image model.
  • Check OPENAI_API_KEY and OPENAI_BASE_URL from ~/.zshrc, then call /models before expensive batches.
  • Do not use low quality.
  • Use at least 2048x1152 high for normal 16:9 slides.
  • Use native 3840x2160 high for complex architecture, dense diagrams, or key customer-facing pages.
  • Never upscale a local 2K image into 4K and call it final.
  • Final pages must be full-page PNGs. Do not build final visual content with local HTML/CSS/Canvas/PPT elements.
  • Do not render fixed page numbers, watermarks, prompt labels, task requirements, internal notes, or draft labels.

If a model returns unexpectedly small images below 2K without user permission, stop batch generation and warn that the provider/model route may be degraded.

If the same repair task returns response too large three times via different attempts, stop and create a handoff prompt instead of retrying indefinitely.

Local Repair Rules

For local fixes such as removing page numbers, repairing text, or changing a small region:

  • Use gpt-image-2 image edit based on the source image.
  • Do not cover bad regions with local translucent masks, solid blocks, pasted text, or screenshots.
  • Repair the needed region with the model, then composite only that region back into the source.
  • Verify dimensions, RGB/opaque output, no residual old text, and pixel-identical unchanged regions.
  • Stage outputs under workspace/runs/\x3Ctask>/ or workspace/tmp/\x3Ctask>/; overwrite final PNG only after review.

For title-style repairs, use a cropped title-header reference instead of a whole slide reference, to avoid reference-slide content leaking into the target page.

Prompt Discipline

Separate visible slide text from design instructions.

Use this prompt shape:

STRICT VISIBLE TEXT ONLY:
customer-visible short phrases only.

DESIGN BRIEF - DO NOT RENDER AS TEXT:
fragmented design instructions.

LAYOUT BRIEF - DO NOT RENDER AS TEXT:
layout and hierarchy.

VISUAL STYLE - DO NOT RENDER AS TEXT:
color, texture, title style.

CONTENT EMPHASIS - DO NOT RENDER AS TEXT:
business meaning to emphasize.

FORBIDDEN VISIBLE TEXT:
describe categories of forbidden text; avoid repeating long forbidden phrases.

Never let the slide show prompt field names, task wording, internal discussion, review notes, or "this page should..." language.

Visual Direction

Default to a premium enterprise/internet style:

  • Light background, blue-white base
  • Small accents in gold, orange, teal/green, or purple
  • Clean, credible, restrained, technical but not flashy
  • Unified left-aligned content-page title system
  • Large readable diagrams/screenshots on image-heavy pages

Avoid cyber neon, heavy dark backgrounds, strong glow, cheap 3D stacks, one-note blue/purple gradients, dense grid backgrounds, oversized titles, and template-like layouts.

PPTX Policy

PPTX is a packaging artifact. It should contain only one full-slide PNG per page.

Do not rebuild PPTX after every small PNG change. For ongoing page edits, generate PNGs and contact sheets first, then ask whether to build PPTX.

When building PPTX, verify:

  • Slide count and order
  • One full-page picture per slide
  • No local text boxes, shapes, notes, page numbers, or placeholder remnants

Useful Script

Use scripts/check_png_assets.py \x3Cpng-dir> to inspect PNG dimensions, mode, and alpha. This is a quick check only; still visually inspect contact sheets and key pages.

Usage Guidance
Review this skill before installing if you do not want agents reading or sourcing your shell configuration. Prefer providing OPENAI_API_KEY and OPENAI_BASE_URL through a scoped runtime environment, verify the configured API endpoint, and only allow the optional frontend demo/dev-server workflow when you explicitly need it.
Capability Tags
requires-sensitive-credentials
Capability Assessment
Purpose & Capability
Generating PPT PNGs, repairing slides, checking image dimensions, and optionally assembling PPTX decks fit the stated purpose. The optional frontend demo guidance expands the workflow, but it is disclosed as a possible deliverable and scoped to the project workspace.
Instruction Scope
The skill explicitly instructs agents to check OPENAI_API_KEY and OPENAI_BASE_URL from ~/.zshrc and provides a command that sources that file before calling /models. API access is expected for image generation, but sourcing a personal shell config is overbroad credential-adjacent handling.
Install Mechanism
No install hooks, package installs, or dependency declarations were found. The included Python script only inspects PNG dimensions and alpha mode.
Credentials
Network calls to the configured OpenAI-compatible endpoint and local file outputs under workspace/ are coherent with the skill. The process-list check and optional local dev server are disclosed, but users should run them only when needed.
Persistence & Privilege
The skill does not create hidden persistence, but it uses sensitive API credentials and may start a dev server for optional prototypes. Credential use is not hidden, yet the ~/.zshrc sourcing approach lacks tight scoping.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ppt-produce-bycodex
  3. After installation, invoke the skill by name or use /ppt-produce-bycodex
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of ppt-produce-bycodex skill for generating and managing enterprise application solution presentations. - Supports full-page PNG slide creation, repair, and assembly into PPTX decks using imagegen/gpt-image-2 models. - Enforces flexible, structured workspace and strict intake workflow for consistent output and asset control. - Includes detailed prompt discipline, visual direction standards, and explicit image/PPTX generation rules. - Provides local repair guidance and sample scripts for asset quality checking.
Metadata
Slug ppt-produce-bycodex
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is ppt-produce-bycodex?

Create and iterate enterprise application software solution PPTs as full-page PNG slides and optional PPTX decks. Use when the user wants to start, design, g... It is an AI Agent Skill for Claude Code / OpenClaw, with 48 downloads so far.

How do I install ppt-produce-bycodex?

Run "/install ppt-produce-bycodex" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is ppt-produce-bycodex free?

Yes, ppt-produce-bycodex is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does ppt-produce-bycodex support?

ppt-produce-bycodex is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created ppt-produce-bycodex?

It is built and maintained by AGICaptain (@javastarboy); the current version is v1.0.0.

💬 Comments