Image-to-Editable-PPT Skill
Convert slide images, PDFs, and image-based PPTX files into editable PowerPoint decks.
Evidence: README explicitly requires full access and describes OCR, image generation/editing, file I/O, subagent dispatch, but no least-privilege option; user confirmation: README mentions AI asks for OCR token, but not all external calls; data flow transparency: README details input normalization, output structure, steps, but no data flow diagram; sensitive data: API keys stored in user-level config with masking, but no encryption; dependency security: no dependency list or vulnerability scan; external effects: third-party APIs called, but data scope not specified; rollback: no rollback mechanism; source attribution: asset provenance mentioned, but not all outputs. Deductions: lack of least privilege, user confirmation, dependency security, rollback evidence.
Evidence: README and tests show CLI design and coverage for concurrency, formula rendering, backend selection, but no full test results; dependency availability: external OCR token and image backend required, but no availability guarantees; failure messages: error handling mentioned, but limited examples. Deductions: dependency availability not fully addressed, failure message evidence limited.
Evidence: README clearly defines use cases (images, PDFs, image-based PPT to editable PPT), distinguishes single-page vs multi-page; capability boundaries clear (no new PPT generation, complex visuals as assets); trigger precision: usage examples provided; environment fit: requires skill loading, file I/O, CLI execution, but not detailed agent compatibility. Deductions: environment fit details insufficient.
Evidence: README provides detailed information architecture (repo structure, output structure); install notes simple; naming stable (CLI commands, file naming); examples and FAQ via docs link; known limitations listed; MIT license; CHANGELOG.md and versioning present; maintenance responsibility: support channels provided, but maintainer identity not explicit. Deductions: maintenance responsibility not explicit.
Evidence: Output is editable PPTX, README shows conversion examples; marginal value: addresses image-to-editable PPT need, but high cost (token consumption); cost-benefit: README warns high cost, recommends Pro users, but no specific cost data. Deductions: cost-benefit evidence insufficient.
Evidence: README claims (conversion quality, cost) supported by examples and warnings, but no reproducible verification steps; cross-source: external article and docs mentioned, but no independent verification; fact vs inference: known issues vs expected outcomes distinguished, but some claims (e.g., 'not 100% replication') are inferences. Deductions: verification evidence insufficient.
- This skill requires full access and may perform OCR, image generation/editing, file I/O, subagent dispatch; ensure trusted environment.
- Third-party API credentials (e.g., OCR token, API key) are stored in user-level config; ensure config file permissions are secure.
- Conversion cost is high and may consume significant tokens; use cautiously based on actual needs.
What does this agent do, and when should you use it?
image-to-editable-ppt-skill is a Codex skill for converting slide images, PDFs, and image-based PPTX files into editable PowerPoint presentations. It normalizes input into per-page tasks and rebuilds them as .pptx: readable text is restored as native text boxes, simple geometry as PowerPoint shapes, and complex visuals are kept as separate image assets with source records. The skill uses a multi-agent collaborative workflow, supporting single-page and multi-page inputs, with multi-page inputs dispatched to page workers for parallel processing. It automatically installs the editppt CLI tool and prefers Codex's built-in image_gen.imagegen for image generation/editing, falling back to third-party APIs when needed. Text correction relies on Baidu PaddleOCR-VL, requiring a free Token. The skill recommends running with full access permissions in Codex due to long execution times and automated steps. Output is always a .pptx file, preserving original slide notes.
The skill performs the following operations: it creates a dedicated task directory and normalizes input into pages/page_NNN/source.png files; based on page count, single-page inputs are rebuilt locally by the main agent, while multi-page inputs are dispatched to page workers according to max_concurrent_pages; page rebuilders complete page reconstruction, self-check, and corrections, creating manifests; finally, the main agent uses editppt run finalize to rebuild the final .pptx in page order, copying slide notes and running deck validation. Image generation and editing prefer Codex's built-in image_gen.imagegen; if unavailable or error, it falls back to the editppt image CLI (using Codex OAuth or OpenAI-compatible API). Text correction calls Baidu PaddleOCR-VL to generate per-page text annotations (bounding boxes, font sizes, groups), improving text restoration accuracy. The output directory structure includes input, final, pages subdirectories, and the final .pptx file.
- Convert a single slide image to an editable PowerPoint for adjusting text and element positions.
- Convert multiple images or multi-page PDFs into a single multi-page .pptx file.
- Convert image-based PPT pages into a more easily editable .pptx, preserving original slide notes.
- Replicate a single-page visual design while keeping text editable.
- Compare source images with output pages to locate missing text, misalignments, or asset gaps.
What are this agent's strengths and limitations?
- Restores text as native text boxes, preserving editability;
- Supports multiple input formats (images, PDFs, image-based PPTs);
- Parallel processing of multi-page inputs via page workers improves efficiency;
- Automatically installs required CLI tools, reducing manual setup.
- Dependent on the Codex environment and not directly usable on other platforms;
- Recommended to run with full access permissions; otherwise frequent approval interruptions may occur;
- Complex visual elements (e.g., photos, illustrations) are only kept as separate image assets, not guaranteed internally editable;
- High conversion cost; a 10-page PPT may consume significant tokens and up to 5 hours of quota.
How do you install or deploy this agent?
To install the skill in a Codex environment, use the following command:
Install the image-to-editable-ppt skill from https://github.com/ningzimu/image-to-editable-ppt-skillAfter installation, the skill automatically installs the editppt CLI tool. To enable text correction, obtain a free Access Token from Baidu AI Studio at https://aistudio.baidu.com/account/accessToken. If a third-party image API fallback is needed, provide the service details and the AI will configure credentials in the user-level config file ~/.editppt/config.yaml (Windows: %USERPROFILE%\.editppt\config.yaml).
How do you use this agent?
In Codex, use the $image-to-editable-ppt directive with image, PDF, or .pptx file paths, for example:
$image-to-editable-ppt Convert this image to an editable PPT.
$image-to-editable-ppt Convert <path-to-deck.pdf> to an editable PPT.The skill automatically handles task directory creation, page worker dispatch (using multi-agent mechanisms), OCR correction, image asset generation, and final .pptx generation. The output file is located in the final directory.
How does this agent compare with similar options?
Unlike codex-ppt-skill, which generates new PPTs from articles, reports, outlines, or ideas, this skill focuses on converting existing images/PDFs/image-based PPTs into editable .pptx files.