Six months of daily Claude Code sessions and I kept reaching for the same three moves: ask a question, get code, iterate on the output. My velocity went up. My range didn't.
The thing I was missing is measurable. Anthropic published a study in February 2026 across 9,830 Claude conversations, classifying 11 distinct collaboration behaviors — things like delegating multi-step tasks, exercising discernment on outputs, or co-constructing plans rather than just querying for results. Most users cluster hard in two or three behaviors and never touch the rest. I wanted to know which ones I was avoiding.
So I built skill-tree: it reads your Claude Code session history, classifies your messages against those same 11 behaviors, and assigns you one of seven archetype cards (rendered as tarot cards using curated museum art — live example at skill-tree-ai.fly.dev/fixture/illuminator). That part is diagnostic. The part I actually care about is the growth quest.
The quest logic has one rule: it never picks a behavior you already do. It finds the 11-behavior slot you've touched zero times or close to it, and surfaces that as your target for the next session. The anti-local-optimum rule. If you're already good at iterative critique, doing more iterative critique isn't growth — it's just confirming what you know. The quest forces you sideways.
The quest persists via a SessionStart hook in ~/.skill-tree/, so Claude Code surfaces it at the top of your next conversation without you having to remember.
Install inside 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, and as a zip plugin for Cowork.
The full analysis — classifier, archetype assignment, narrative, rendered card — runs in 30–60 seconds and returns a stable URL.
Top comments (0)