Claude Skills by marsmike

Scaffold v2 `.slide.dsl` layout skeletons for a brand pack from its claude-design HTML. Use when adding new layouts.

Iteratively improve a brand pack's `.slide.dsl` layouts to match a source PPTX. Runs the verify-loop, then fans out one sub-agent per layout above threshold so each slide is analysed and edited in parallel.

Build or polish a brand-compliant PowerPoint deck via the v2 DSL pipeline. Use when the user asks to create a deck, make a presentation, or polish a rough .pptx.

Use when visualizing systems, architectures, or concept flows — brand-aware Excalidraw authoring via compact DSL

Use when authoring custom data viz that doesn't fit a predefined chart layout — brand-aware SVG infographics via compact DSL.

Generate AI images via Replicate or Gemini. Use to create illustrations, photos, or graphics from a text prompt.

Convert text to speech with ElevenLabs voiceover. Use when generating speech audio or browsing available voices.

Iteratively improve a feinschliff diagram skill: loop measure, mutate one thing, keep or revert, consolidate, scoring generated output with `feinschliff-builder eval`. Rides the built-in /goal. Use when asked to autoloop against its evals.

Use when authoring a CLI session recording for the cli-recorder skill. Translates a user's intent ("video of installing kubectl") into a valid recipe.toml through interactive Q&A.

Use when creating programmatic videos, animated explainers, or visual compositions with Remotion.

Editing a pre-recorded video into a brand-themed cut — authors edit_plan.json with overlay beats, captions, and audio score; probe/transcribe/lint/preview/final pipeline. Use when the user provides footage and asks to edit, enhance, or caption it.