Anthropics February 2026 study classified 11 observable collaboration behaviors across 9,830 Claude conversations. The researchers called it the AI Fluency Index. I wanted to run the same classification on my own session history — not to benchmark against other users, but to find out which of the 11 behaviors I never touched.
The result is skill-tree. It reads your Claude Code or Cowork session files, runs a remote classifier against the same 11-behavior taxonomy from Dakan and Feller's 4D AI Fluency Framework (three observable axes: Description, Discernment, Delegation — the fourth, Diligence, doesn't show up in chat logs), and assigns you one of seven archetype cards rendered as tarot cards with curated museum art.
The archetypes — Illuminator, Navigator, Alchemist, and four others — aren't just cosmetic. Each maps to a distinct behavioral cluster. The Illuminator front-loads context and over-explains; the Navigator iterates tightly and discards failed directions fast. Seeing which card you land on tells you something your git history doesn't: not what you shipped, but how you collaborated to get there. You can see a live rendered example at skill-tree-ai.fly.dev/fixture/illuminator.
The growth quest piece is what I actually find useful day-to-day. After classification, the tool picks one behavior from your blind spots and pins it as a quest via the SessionStart hook, so it shows up at the top of your next Claude Code session. The state persists in ~/.skill-tree/
between sessions.
The full pipeline — find session files, extract user messages, classify remotely via Claude Haiku on Fly.io, assign archetype, synthesize narrative, render the card, return a stable URL — runs in 30 to 60 seconds.
Install in Claude Code:
claude plugin marketplace add robertnowell/ai-fluency-skill-cards
claude plugin install skill-tree-ai@ai-fluency-skill-cards
Also works in Cowork via skill-tree-ai.zip
, or as an MCP server (npm package skill-tree-ai
) for Cursor, VS Code, and Windsurf.