Case study: CC Adoption Wave Explorer

Problem. Claude Code ships multiple releases a week. Every release can carry breaking changes for a plugin with 100+ skills and 37 agents — hooks that stop firing, permission semantics that shift, worktree isolation that tightens. Reading changelogs manually does not scale, and falling one version behind costs hours of debugging.

Approach. An automated pipeline (described in How we adopt every Claude Code release) watches upstream releases, snapshots each changelog, and runs an LLM triage that files one scored GitHub issue per change. For the July 2026 wave that meant 37 issues across CC 2.1.210 → 2.1.212, each with a category (breaking / new command / new field), a 0–20 gap score, and the list of affected skills.

The explorer below is the dashboard that wave produced. Filter by version, category, gap score, or affected skill; click cards to build a work batch; the prompt bar composes the triage instruction that goes back into a Claude Code session.

What it demonstrates. Chart marks use a CVD-validated categorical palette (6 slots, all accessibility checks passing), stat tiles and stacked bars follow a documented encoding standard, and the whole artifact is one dependency-free HTML file generated inside the working session that triaged the wave.