The Best PPT Skills for Claude Code & Codex in 2026: 8 Options Compared

A PPT skill is a

package that teaches a coding agent — Claude Code, OpenAI Codex, OpenClaw, Cursor — how to produce PowerPoint files. In 2026 the ecosystem splits into three families: code-composed skills that write PPTX programmatically (Anthropic's pptx skill, OpenAI's Slides skill), image-composed skills that render each slide as AI art (codex-ppt-skill, banana-slides, nanobanana-ppt), and hosted skills that call a generation API (2Slides). This comparison covers eight options and which family fits which job.

Full disclosure up front: 2Slides is our product, and it appears in this list. Everything else here is assessed from public documentation, repos, and published behavior, and we say plainly where other tools are the better choice.

How PPT skills work (60-second primer)

An Agent Skill is a folder of instructions, scripts, and resources an agent loads when the task matches — Anthropic's definition, and OpenAI's is equivalent. For presentations, the skill decides three things: how content is planned (outline first, style contract, per-slide brief), how slides are produced (library code, image generation, or an API call), and how the result is verified (rendering slides back to images to catch overflow is the current best practice). Installation is one command in most agents —

in Codex, or dropping the folder into your skills directory in Claude Code.

The eight skills

1. Anthropic pptx skill (official, Claude)

The same code Claude uses internally for file creation. Builds new decks with PptxGenJS, edits existing ones by unpacking and rewriting slide XML, renders slides to images to check its own work, and ships built-in guidance on color, type, and layout. Best at: editing an existing PPTX in place — no other skill in this list does XML-level edits of decks it didn't create. Weakest at: visual ambition; output is clean corporate-plain.

2. OpenAI Slides skill (official, Codex)

Ships with Codex as a system skill. Writes PptxGenJS, keeps text as text, prefers native PowerPoint charts for data, and runs render/overflow/overlap/font checks before finishing. Best at: chart-heavy internal decks produced entirely inside Codex. Weakest at: brand-matched design; layout quality depends on the code the model writes that session.

3. codex-ppt-skill (open source)

The project that popularized image-based decks in coding agents: plan outline and visual style, generate each slide as a full-bleed image, assemble to

with a local script. Documented to run in Codex, Claude Code, OpenClaw, and Hermes. Best at: art-directed slides from a single style prompt. Weakest at: editability — text is baked into pixels — and per-run consistency.

4. banana-slides (open source)

An AI-native generator built on Google's nano banana pro image model, aiming at what its author calls "Vibe PPT": upload any template image, feed materials, generate from a sentence or an outline, then revise specific regions conversationally. Exports to editable PPT. Shipped rapid releases through July 2026. Best at: template-image-driven visual style and conversational regional edits. Weakest at: enterprise predictability; it's a fast-moving young project. (More on the trend it named: What is Vibe PPT?)

5. nanobanana-ppt skill (open source)

A lighter-weight skill in the same image-composed family, listed across the skill directories: nano-banana image generation per slide, assembly script, minimal configuration. Best at: quick visual drafts and social-format graphics. Weakest at: long documents and data slides.

6. 2Slides skill (hosted — our product)

A

that delegates generation to the 2Slides API from any supporting agent. Two modes: Fast PPT fills professionally designed templates and returns native, editable PPTX (10 credits/page); Nano Banana renders image-designed slides at 1K–4K (100–200 credits/page). Same job can add 30-voice narration and MP4 export. Install: . Best at: consistency across runs, recurring/branded decks, multi-language output (19+ languages), and being the only option here with narration + video on the same pipeline. Weakest at: offline use — it's an API, credits cost money after the free 500, and template-based layouts trade away the fully bespoke look.

7. ChatGPT @Presentations skill (official, ChatGPT app)

Not a coding-agent skill, but users compare them constantly. Attach source material, reference an existing deck or template, and ChatGPT plans, builds, and validates a Google Slides or PowerPoint file. OpenAI's own docs note template matching and advanced formatting may require manual refinement. Best at: everyday users inside ChatGPT with Drive integration. Weakest at: automation — no scripting, chaining, or repo context. (Full review.)

8. DIY: python-pptx + your own SKILL.md

The build-it-yourself path: a short skill file pinning python-pptx conventions, your corporate template, and a validation script. Best at: total control and zero external dependencies. Weakest at: everything you now have to maintain yourself — most teams that start here end up adopting one of the options above once the deck count grows. (Guide: build an AI presentation agent.)

Comparison table

Which should you pick?

• You edit existing decks more than you create them → Anthropic pptx skill; XML-level editing is its unique trick.
• You live in Codex and make occasional internal decks → the official Slides skill is already installed.
• You want maximum visual drama and nobody edits the file after → codex-ppt-skill or banana-slides.
• Decks recur, carry a brand, or ship to clients → hosted generation (2Slides skill) is the only family where output is template-locked instead of model-mood-dependent.
• You're a non-developer in ChatGPT → @Presentations.
• You have unusual compliance constraints → DIY python-pptx keeps everything local.

FAQ

What is a PPT skill?

A PPT skill is an Agent Skill — a folder with a

of instructions, scripts, and resources — that teaches an AI coding agent to produce PowerPoint files reliably. It pins the workflow (plan → generate → verify) so the agent doesn't improvise a new approach every session.

Are PPT skills better than MCP servers for presentations?

They solve different problems: a skill teaches the agent to do the work locally; an MCP server exposes a remote tool the agent calls. Skills win on offline control, MCP wins on zero-maintenance. Full comparison: Agent Skills vs MCP servers.

Do these skills work in Cursor and OpenClaw?

Most community skills and the 2Slides skill document multi-agent support (Codex, Claude Code, OpenClaw, Cursor share the SKILL.md convention). The two official skills are tied to their own agents — Anthropic's to Claude, OpenAI's to Codex.

Can image-based PPT skills produce editable slides?

Mostly no — text is rendered into the image, so edits mean regeneration. banana-slides exports an editable PPT layer and 2Slides' Fast PPT mode is fully native; those are the exceptions in this list.

Where do I discover new skills?

Skill directories emerged in 2026 as the ecosystem's index: awesomeskill.ai and Agent Skills Hub both maintain PPT categories, and OpenAI's

repo documents the official set and installer flow.