After months of daily Claude Code use, I ran my session history through Anthropic's AI Fluency Index taxonomy and found I'd triggered 8 of the 11 observable behaviors — but three had zero occurrences. Not low. Zero.

That's the problem skill-tree is built around. It's not hard to get faster at the things you already do. The trap is that fluency gains feel real even when you're just deepening the same ruts.

What it does

skill-tree installs as a Claude Code plugin, reads your session history, and classifies your messages against the same 11 behaviors Dakan & Feller defined in their 4D AI Fluency Framework — the ones Anthropic used as the population baseline in their February 2026 study of 9,830 conversations. The output is an archetype card (seven archetypes, rendered as tarot cards with museum art) and a growth quest.

The quest is deliberately anti-local-optimum. It doesn't pick a behavior you do occasionally and could refine. It picks the behavior with the lowest occurrence rate in your history — the one you structurally avoid. The logic: if you're already visiting a behavior, you have some feedback loop on it. If you've never triggered it in months of daily use, your mental model probably doesn't include it at all.

Here's what the install looks like:

claude plugin marketplace add robertnowell/ai-fluency-skill-cards
claude plugin install skill-tree-ai@ai-fluency-skill-cards

The 7-step orchestration (find session files → extract user messages → remote classifier → archetype assignment → narrative synthesis → render → return URL) runs in 30–60 seconds and returns a stable hosted URL. Live example at skill-tree-ai.fly.dev/fixture/illuminator.

The growth quest persists via a SessionStart hook so it surfaces at the top of your next session, not buried in a report you close and forget.

Also available as an MCP server (npm install skill-tree-ai) for Cursor, VS Code, and Windsurf.

github.com/robertnowell/skill-tree