Starstruk Logo MarkStarstruk

v2.0.0

commaai/openpilotv2.0.0Mar 4, 2026by zarazhangrui

AI Summary

This release implements a progressive disclosure restructure to reduce context bloat by approximately 89%, transforming the skill from 1,625 lines to a leaner architecture with on-demand loading of supporting files. The update follows best practices from Claude Code's skill documentation and OpenAI's harness engineering approach while maintaining all existing functionality.

Key Highlights

  • Reduced context bloat from 1,625 to 183 lines in core SKILL.md file
  • Implemented progressive disclosure pattern with on-demand loading of supporting files
  • Deduplicated viewport CSS into single canonical file
  • Converted PPT extraction to executable Python script
  • Embedded design aesthetics prompt for consistent output quality

Breaking Changes

  • Requires re-cloning the skill directory to apply changes

New Features

  • Progressive disclosure architecture
  • On-demand loading of STYLE_PRESETS.md during phase 2
  • On-demand loading of viewport-base.css during phase 3
  • On-demand loading of html-template.md during phase 3
  • On-demand loading of animation-patterns.md during phase 3
  • Executable scripts/extract-pptx.py for PPT conversion

Full Release Notes

## What's New

This release completely restructures the skill to reduce context bloat by ~89%, following best practices from [Claude Code's skill docs](https://code.claude.com/docs/en/skills) and [OpenAI's harness engineering](https://openai.com/index/harness-engineering/) approach.

### The Problem

The previous version loaded **1,625 lines** of instructions into Claude's context on every invocation — including CSS code blocks, Python scripts, and HTML templates that were only needed during specific phases. This crowded out the actual presentation content and reduced output quality.

### The Fix: Progressive Disclosure

`SKILL.md` is now a **183-line map** (was 1,098). Supporting files are loaded on-demand only when Claude reaches the relevant phase:

| File | Purpose | Loaded When |
|------|---------|-------------|
| `SKILL.md` | Core workflow and rules | Always |
| `STYLE_PRESETS.md` | 12 curated visual presets | Phase 2 (style selection) |
| `viewport-base.css` | Mandatory responsive CSS | Phase 3 (generation) |
| `html-template.md` | HTML structure and JS features | Phase 3 (generation) |
| `animation-patterns.md` | CSS/JS animation reference | Phase 3 (generation) |
| `scripts/extract-pptx.py` | PPT content extraction | Phase 4 (conversion) |

### Other Changes

- **Deduplicated viewport CSS** — was repeated across both files, now lives in one canonical `viewport-base.css`
- **Embedded design aesthetics prompt** — anti-AI-slop instructions are now always in context
- **PPT extraction is now an executable script** — `scripts/extract-pptx.py` can be run directly
- **Removed ~566 lines of duplication** — total content is 1,059 lines (down from 1,625)
- **Updated README** with new architecture explanation

### Upgrading

```bash
# Re-clone to your skills directory
rm -rf ~/.claude/skills/frontend-slides
git clone https://github.com/zarazhangrui/frontend-slides.git ~/.claude/skills/frontend-slides
```

All existing functionality is preserved — same 12 style presets, same 5-phase workflow, same PPT conversion support.