DEV Community

J Now
J Now

Posted on

Months of Claude Code sessions, no idea if I was improving

I've been using Claude Code daily for months. Fast, yes. But I started noticing I had the same three moves: ask a question, paste in an error, ask for a refactor. I had no idea whether I was actually developing any depth or just getting faster at a narrow groove.

Anthropic published a study in February 2026 (9,830 Claude conversations, 11 observable collaboration behaviors) that gave me a concrete baseline to check myself against. I built skill-tree to run that same classification on my own session history.

Here's what it does:

claude plugin marketplace add robertnowell/ai-fluency-skill-cards
claude plugin install skill-tree-ai@ai-fluency-skill-cards
Enter fullscreen mode Exit fullscreen mode

The 7-step orchestration takes 30–60 seconds: it finds your local session files, extracts user messages, ships them to a remote Claude Haiku classifier on Fly.io, maps results against the 11 behaviors from the AI Fluency Index, assigns one of seven archetype cards, writes a narrative, renders a visualization, and returns a stable URL. The archetype cards are rendered as tarot cards using curated museum art β€” see a live example at skill-tree-ai.fly.dev/fixture/illuminator.

The part I built for myself specifically: a growth quest. After classification, it identifies which of the 11 behaviors are absent or underrepresented in your sessions and surfaces one as a concrete thing to try in your next session. That quest persists across sessions via a SessionStart hook (~/.skill-tree/ in Claude Code).

The behavior taxonomy comes from Dakan & Feller's 4D AI Fluency Framework β€” three axes are observable in chat logs (Description, Discernment, Delegation); the fourth (Diligence) isn't captured from logs alone.

Also available as an MCP server (npm package skill-tree-ai) for Cursor, VS Code, and Windsurf if you're not on Claude Code.

https://github.com/robertnowell/ai-fluency-skill-cards

Top comments (0)