From catalan-skills
Assess your AI fluency using Anthropic's 4D framework (Dakan, Feller & Anthropic, 2025). Scans Claude Code sessions, runs LLM-based behavior classification on all messages, asks a self-assessment questionnaire for 6 unobservable behaviors, and generates a visual HTML report with scores and actionable feedback. Use when "assess fluency", "AI fluency", "fluency report", "fluency assessment", "4D framework", or "how AI fluent am I".
How this skill is triggered — by the user, by Claude, or both
Slash command
/catalan-skills:ai-fluency-assessmentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Assess a user's AI fluency based on Anthropic's 4D AI Fluency Framework (Dakan, Feller & Anthropic, 2025). The framework has 4 Competencies, 12 Sub-competencies, and 24 Behaviors. The `assess.py` script classifies 18 behaviors from conversation data using an LLM binary classifier (matching Anthropic's own methodology). The remaining 6 behaviors require a self-assessment questionnaire.
Assess a user's AI fluency based on Anthropic's 4D AI Fluency Framework (Dakan, Feller & Anthropic, 2025). The framework has 4 Competencies, 12 Sub-competencies, and 24 Behaviors. The assess.py script classifies 18 behaviors from conversation data using an LLM binary classifier (matching Anthropic's own methodology). The remaining 6 behaviors require a self-assessment questionnaire.
First, ask the user:
~/.claude/projects/ — is that correct, or should I scan a different directory?"Then run assess.py to scan and classify all 18 observable behaviors. The script lives next to this SKILL.md — substitute the actual path when running.
# Default: scan ~/.claude/projects/, classify with Claude Haiku
python3 /path/to/skills/ai-fluency-assessment/scripts/assess.py \
--output-dir .ai-fluency --max-sessions 6000
# Custom sessions directory (e.g., exported sessions, another machine's data)
python3 /path/to/skills/ai-fluency-assessment/scripts/assess.py \
--sessions-dir <PATH> --output-dir .ai-fluency --max-sessions 6000
# Fast/free fallback (regex only, 11 behaviors, less accurate)
python3 /path/to/skills/ai-fluency-assessment/scripts/assess.py \
--output-dir .ai-fluency --max-sessions 6000 --regex-only
Use --max-sessions 6000 to capture all sessions. Many session files are subagent files with no user messages, so even with thousands of files the scan is fast.
LLM classification (default) requires the anthropic SDK and API credentials. The script auto-detects Anthropic Foundry credentials from environment variables (ANTHROPIC_FOUNDRY_API_KEY), or uses ANTHROPIC_API_KEY for direct API access. Cost: ~$2-5 per full assessment with Claude Haiku.
Output files:
.ai-fluency/evidence.json — all extracted user messages with project metadata.ai-fluency/analysis.json — per-behavior match counts, sample messages, top projectsAsk the user the following 6 questions for behaviors that have no observable signal in conversation data. Rate each 1-5 (1=Never, 5=Always).
Present them in conversation — don't run an interactive CLI.
Delegation:
Discernment:
Diligence:
After collecting responses, save them:
python3 -c "
import json
from pathlib import Path
responses = {3: R3, 4: R4, 18: R18, 19: R19, 20: R20, 24: R24}
evidence_path = Path('.ai-fluency/evidence.json')
with open(evidence_path) as f:
evidence = json.load(f)
evidence['questionnaire'] = {str(k): v for k, v in responses.items()}
with open(evidence_path, 'w') as f:
json.dump(evidence, f, indent=2, default=str)
print('Questionnaire saved')
"
Replace R3, R4, etc. with the user's actual integer ratings.
Read .ai-fluency/analysis.json for the 18 LLM-classified behaviors and .ai-fluency/evidence.json for the questionnaire responses. Score each behavior 1-5:
Scoring criteria:
For LLM-classified behaviors (B1-B2, B5-B17, B21-B23): Base scores on LLM match counts, quality of sample messages, consistency across projects, and range/variety within the behavior.
Important: High frequency alone does NOT guarantee a high score. Assess quality and range, not just quantity.
For questionnaire-only behaviors (B3, B4, B18, B19, B20, B24): Use the self-assessment rating directly.
Generate a self-contained HTML report at .ai-fluency/fluency-report.html following the report spec for design system, layout, and formatting specifications.
open .ai-fluency/fluency-report.html
See the framework reference for the full 24-behavior reference table.
npx claudepluginhub catalan-adobe/skills --plugin catalan-skillsGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.