- Home
- Skills
- Prompt & Skill Engineering
- AI Fluency Score
Works with the AI Agents you already use
AI Fluency Score
Quantifiable AI collaboration scoring for Claude Code users based on the Anthropic 4D Fluency framework.
Free
AI Fluency Score
Example session with this skill installed
python3 "${CLAUDE_SKILL_DIR}/driver.py" score
- Read your context and instructions
- Compiled the ai fluency score
Successfully analyzed 142 sessions from the local evidence pack.
AI Fluency Score (4D Framework)
| Dimension | Score | Raw | Primary Signal |
|---|---|---|---|
| Delegation | 3/5 | 3.25 | runway median (tool calls/instruction): 6 |
| Description | 2/5 | 1.80 | instruction length median: 38 chars |
| Discernment | 4/5 | 4.10 | steering latency (agent msgs, median): 1 |
| Diligence | 3/5 | 2.90 | sessions where you asked for a check %: 40 |
Strongest: Discernment
Weakest: Description
Analysis: You have a safe asymmetry profile. Your high Discernment helps catch errors, but your weak Description (sub-40 character median) suggests you are relying on the agent to guess your intent, which increases steering latency over time.
Next steps
- Run
python3 "${CLAUDE_SKILL_DIR}/driver.py" cadenceto see when you should re-evaluate. - Increase your initial instruction detail to move the Description score toward a 4.
- Use the
/modelswitch more frequently to improve Delegation via routing.
Connects securely to your tools. The creator never sees your data.
About this skill
You are collaborating with an AI, but are you doing it well? AI fluency is the gap between simply using an LLM and mastering it as a professional leverage point. This tool analyzes your local Claude Code transcripts to provide a data-driven audit of your prompting habits and delegation logic.
What it does
- Quantifies AI Fluency by scoring your performance across four dimensions: Delegation, Description, Discernment, and Diligence.
- Extracts Sub-signals from transcripts, such as tool-call runway medians, model routing frequency, and manual verification triggers.
- Determines Re-run Cadence using statistical confidence intervals to tell you exactly how many sessions you need before a score update is statistically significant.
- Redacts Sensitive Data by filtering out credentials, keys, and PII from transcripts before they reach the evidence pack.
- Identifies Risk Asymmetry by flagging dangerous habit pairings, like high Discernment with low Description, which leads to "polished mistakes."
How it works
- Initialize the environment and verify local transcripts using the
checkcommand. - Extract session data into a secure, local evidence pack that summarizes your interaction patterns.
- Score your performance to see 1-5 ratings on the 4D framework and identify your strongest and weakest traits.
- Analyze your improvement window to see the recommended number of days between evaluations.
Frameworks & tools
This skill is built for Claude Code and runs locally using Python 3.9+. It processes standard .jsonl transcripts stored in ~/.claude/projects.
Why this beats prompting it yourself
Manual self-reflection is biased and lacks the granularity to track tool-call runway or routing percentages. This tool uses a deterministic rubric and statistical confidence math to provide an objective baseline that no "how am I doing?" prompt can replicate.
Use cases
- Auditing your own developer productivity and identifying where you are micro-managing the agent.
- Measuring the impact of new prompting techniques over a 30-day window.
- Ensuring your local AI usage habits are not trending toward "autopilot" errors.
Known limitations
Only supports Claude Code transcripts; other agents or IDE extensions are not compatible. Requires local transcript history to exist before scoring can begin.
Only supports Claude Code. Requires local transcripts in JSONL format. Requires Python 3.9+. Scores are based on fixed proxies, not human intuition.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 18 days ago
- Free with an Agensi account
Requires