Slides Grab Presentation Mode Router
End-to-end slide generation inside Claude Code or Codex: routes each request to an HTML or image-native pipeline, then runs plan, design, and export stages to PDF, PNG, and more.
Evidence shows a pure routing skill with no destructive commands, explicit per-stage user approval, a design gate before export, and asset-localization rules banning remote image URLs; however the default image provider reuses local ~/.codex/auth. credentials against an 'unsupported private Codex backend' — disclosed with fallbacks, but credential data-flow transparency and rollback notes remain incomplete, so no high score.
SKILL.md and the reference doc are self-consistent with clear routing, stage order and gate logic; the repo contains substantial test suites (design-gate, codex-imagen, design-md-parser), but no test-running CI workflow appears in the provided evidence (only pages/publish), and a core image path depends on a private backend that may break without notice, with PPTX/Figma flagged unstable; static review cannot verify key-path reproduction, so scored 9.
Scenarios are clearly partitioned (signal table for HTML vs image-native, HTML default), non-fit boundaries (experimental exports) are declared, and trigger semantics are reasonably precise; but no Chinese-language support is declared, and image generation / video fetching depend entirely on overseas services (Codex login, OpenAI, YouTube/yt-dlp), a real mainland-China reachability risk — deducted here.
Docs are well layered (router SKILL.md + reference + sub-skill links), MIT license is explicit, attribution is honest (ppt_team_agent credit), and experimental limitations are well disclosed; but the skill itself carries no version/changelog, install guidance sits in the Korean README, and maintenance responsibility is only inferable from the author field — deducted.
The declared Plan→Design→Export workflow is structurally complete with full command coverage and marginal value over manual slide-making; but the primary outputs (PDF/PNG/PPTX) are not verified as directly usable in the evidence, the author flags PPTX/Figma as needing manual cleanup, and the image path depends on an unstable backend — scored 6 under static calibration.
Primary evidence is author-authored docs plus promotional README material (demo gif, showcase link); the repo does contain auditable test source covering design-gate, import-injection safety (UNTRUSTED boundaries) and image contracts, but there is no third-party execution evidence or test-running CI in the provided files, limiting independent reproduction — scored 4.
- Default image generation reuses local Codex login credentials against an unsupported private backend; review the credential data flow yourself and use the documented openai/nano-banana or local-download fallbacks.
- PPTX and Figma exports are experimental/unstable; outputs may need manual cleanup and should not be used for final deliverables without review.
- Image generation and video fetching depend on overseas services (OpenAI, YouTube/yt-dlp) that may be unreachable from mainland-China networks.
- This assessment is a static source read only; no tests were executed and no test-running CI evidence was present.
- This skill is a router and requires the slides-grab-plan/design/export sub-skills to be installed.
What does this skill do, and when should you use it?
slides-grab is a presentation-generation skill for Claude Code and Codex that acts as a mode router: it reads the user's intent and dispatches the request to either a semantic HTML pipeline or an image-native pipeline. Both pipelines share the same planning (Stage 1) and export (Stage 3) stages; PDF and per-slide PNG are the preferred exports, while PPTX and Figma are labeled experimental and unstable. The skill enforces stage ordering and a design gate, so export only happens after a Proceed verdict. It is one of 7 skills bundled in the NomaDamas/slides-grab repository.
Reads the user request and picks a delivery mode: editable text, searchable content, accessibility, charts, or diagrams route to slides-grab-html; explicit priority on matching an existing corporate template routes to slides-grab-image, defaulting to HTML when in doubt. Planning uses the slides-grab-plan skill and records mode: html or mode: image-native in slide-outline.md metadata. Export uses the slides-grab-export skill, running slides-grab pdf, slides-grab png (including card-news 1:1 mode), and optionally slides-grab convert (PPTX) or slides-grab figma. Export is gated on a Stage 2 design-gate Proceed verdict; missing, stale, or contract-violating receipts block pdf/convert/figma. Charts default to Chart.js in-slide canvas rendering with slides-grab validate catching empty canvases; complex diagrams default to tldraw-generated assets; bespoke imagery defaults to codex-imagen (reusing the local Codex login, no API key needed).
- An engineer who wants a topic or set of files turned into a full deck, with Claude Code or Codex handling planning through export automatically
- A team handed a corporate PPTX/PDF template who need new slides to visually match it closely
- A social-media operator producing Instagram-style 1:1 card-news image sets
- An analyst embedding data charts in slides who wants automatic validation that no canvas renders blank before export
- A presenter who needs per-slide 2160p PNGs or a searchable-text PDF rather than a flat raster
What are this skill's strengths and limitations?
- Automatically routes between semantic HTML and image-native pipelines; HTML mode yields editable, searchable, accessible slides
- Enforced stage order and design gate give CLI-level hard constraints on export quality
- Rich export options: PDF (capture/print modes), per-slide 2160p PNG, card-news 1:1 output
- Honest expectation-setting: PPTX and Figma are explicitly flagged experimental
- PPTX and Figma exports are experimental/unstable, with possible layout shifts and manual cleanup
- Heavy runtime dependencies: Node.js >= 20, Playwright Chromium, and the full slides-grab CLI
- Must be installed together with several sibling skills from the repo; this router alone cannot function
- Default image generation relies on codex-imagen's unsupported private Codex backend, which the docs warn may break at any time and requires an account entitled to image generation
How do you install this skill?
No clone needed: npm install slides-grab && npx playwright install chromium, then npx slides-grab install-skills --target all --scope user to install shared skills plus runtime adapters into Codex and Claude Code. Alternatively install via Vercel Agent Skills: npx skills add ./node_modules/slides-grab -g -a codex -a claude-code --yes --copy --full-depth. Requires Node.js >= 20. The source does not document installing this router skill alone; at runtime it depends on sibling skills from the same repo: slides-grab-plan, slides-grab-html, slides-grab-image, and slides-grab-export.
How do you use this skill?
Trigger it in Claude Code or Codex with natural language, e.g. "make slides about X" (HTML mode) or "make slides matching this existing template" (image-native mode). The skill confirms the mode, runs planning and design, then exports once the design gate returns Proceed. Common exports: slides-grab pdf --slides-dir decks/my-deck --output my-deck.pdf; slides-grab png --slides-dir decks/my-deck --slide-mode card-news --resolution 2160p. Multi-deck projects live under decks/<deck-name>/.
How does this skill compare with similar options?
The repository describes itself as an open-source alternative to Claude Design; the README also credits Builder Josh's ppt_team_agent as its foundation. Unlike tools that generate PPTX directly, it uses HTML/CSS as the slide medium and supports pointing at a region in a browser editor for in-place agent edits.