After months of daily Claude Code sessions I noticed a pattern: I was fast, but my prompts all looked the same. Lots of delegation ("write this function"), almost no discernment ("critique this approach before we proceed"). I had no data to confirm that, just a nagging feeling.
Anthropic published a study in February 2026 — 9,830 Claude conversations, 11 observable collaboration behaviors — and I wanted to run the same classification on my own sessions. So I built skill-tree-ai to do exactly that.
The tool reads your Claude Code or Cowork session history, runs a remote classifier (Claude Haiku on Fly.io) against all 11 behaviors from the AI Fluency Index, then assigns one of seven archetype cards. The archetypes are rendered as tarot cards with curated museum art — Illuminator, Alchemist, Navigator, and four others. Each card includes a skill radar showing where your behavior clusters relative to the three observable axes: Description, Discernment, and Delegation. The fourth axis, Diligence, comes from Dakan & Feller's 4D framework but isn't extractable from chat logs, so it's excluded.
The part I use most: the growth quest. After classification, the tool picks a specific behavior from the 11 that you haven't touched and surfaces it as a concrete prompt pattern to try next session. The quest persists via a SessionStart hook — ~/.skill-tree/ in Claude Code, $CLAUDE_PLUGIN_ROOT/.user-state/ in Cowork because $HOME is ephemeral there.
Whole pipeline takes 30–60 seconds: find session files → extract user messages → remote classifier → archetype assignment → narrative synthesis → render → stable URL. You can see a live fixture at skill-tree-ai.fly.dev/fixture/illuminator.
Install in Claude Code:
claude plugin marketplace add robertnowell/ai-fluency-skill-cards
claude plugin install skill-tree-ai@ai-fluency-skill-cards
Also available as an MCP server (npm install skill-tree-ai) for Cursor, VS Code, and Windsurf.
Top comments (0)