How to use the Nature Image2PPT agent skill (nature-image2ppt)

What an agent skill is and what this one makes the agent do

An agent skill is a folder of instructions and scripts that an AI agent can load when a request matches the skill's description. This one is scoped to reconstruction: it restores an existing visual page as an editable PowerPoint file rather than writing a new presentation from notes.

The skill follows a staged workflow. The agent checks its dependencies, optionally reads text with OCR, splits each page into regions, writes a manifest per page, and then assembles and checks the final .pptx. Simple measured objects such as text boxes, cards, circle nodes and connectors are rebuilt natively. Complex artwork is kept as cropped image assets that you can replace.

• Input: slide images, screenshots, scanned PDFs or image-only PPTX files.
• Output: a .pptx where text, shapes and connectors are separate objects, plus per-page manifests, text hints, render previews and QA reports.
• Not covered: writing a new deck from research notes or an outline.

Install the skill from the ZIP

Download the ZIP of the skill from this page. Then choose one of two routes.

For a personal setup, unzip the folder into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

The skill's SKILL.md expects Python 3.10 or later and the packages listed in requirements.txt. If you use a dedicated environment, the agent is told to use the environment's Python instead of the system one.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check requirements before the first run

The skill runs a preflight command called doctor. If it reports a missing dependency, the source says to install only that reported dependency and rerun doctor before continuing.

OCR is optional. With a Baidu AI Studio PADDLE_OCR_TOKEN, the skill uses cloud OCR. Without it, the skill falls back to a local builtin-ink mode that measures text geometry but does not recognise characters. Image generation for asset work is also optional and needs either Codex OAuth for GPT Image models or an OpenAI-compatible API key and endpoint.

• Python 3.10 or later and requirements.txt packages (including PIL and defusedxml).
• Optional: PADDLE_OCR_TOKEN for cloud OCR.
• Optional: image backend credentials (Codex OAuth, or an OpenAI-compatible API key and endpoint).
• Comfort running Python tools and installing dependencies.

Trigger the skill with a natural request

You do not need a special command to start. Describe what you have and what you want back. The skill's description names Chinese requests such as 图片转可编辑PPT and 截图还原PPT, and diagram reconstruction.

Useful requests say what the source is, what must stay editable, and any limits such as page range or offline handling. Requests asking for a brand-new deck from notes do not match this skill.

• "Reconstruct these slide screenshots as a PowerPoint with individually editable elements."
• "Convert this scanned PDF into an editable PPTX while preserving the original layout."
• "Repair this image-only PPT so the flowchart and arrows can be edited."
• "图片转可编辑PPT：把这些截图还原成PPT，保留流程图。"

What the agent produces and how it checks the work

The agent prepares one run, works page by page, and gates each page with a build step and a QA script. Pages are written into their own directories, and the final deck is assembled from a run-level manifest. The final file is built to a temporary file and only published after a successful build.

Arrows are represented as one connector or one Arrow shape, not a line plus a triangle. Formulas are a hard gate: if the formula engine fails, the page stays failed unless you explicitly approve that exception.

• Per-page manifest, text hints, render preview and QA report.
• Final PPTX validation, plus a list of complex visuals kept as replaceable image assets.
• Speaker notes from source PPTX files are preserved, according to the project README.

Safety notes: scripts and network access

This skill runs scripts and uses the network. Image generation and OCR may send slide content to third-party services. If your slides are sensitive or confidential, choose an offline path and do not configure cloud OCR or image backends.

The skill's rules keep each page's output inside its own directory and the final output inside the run directory. Any file that tries to escape those boundaries is treated as a hard failure.

• Respect an offline-only choice; the agent is told to honour it.
• Cloud OCR and image generation can upload the page image and prompts.
• Review the run directory and page directories after each job.

Tips for better results

Give the cleanest source you have. A sharp screenshot or a high-resolution scan gives the agent more to measure. Tell the agent which parts must stay native and which complex illustrations can stay as images.

Expect the fallback to need your review. Without OCR the text is positioned by geometry, not read as characters, so check wording on each page.

• Send one page or a small batch first to learn how the output looks.
• Ask for diagrams and flowcharts explicitly so connectors and arrows are rebuilt as objects.
• Keep the project's README and SKILL.md nearby if you need to adjust settings.