AI Usage Audit — Monthly Retrospective & Insights

AI Usage Audit — Monthly Retrospective & Insights

Purpose

Turn your AI conversation history into a mirror: see what you're actually using AI for, where you're efficient, where you're wasting time, and where collaboration friction is highest. It's essentially a Sprint Retro applied to human-AI collaboration — a data-driven approach to continuously improving how you work with AI.

When to Trigger

Prerequisites

This skill depends on recent_chats and conversation_search tools to pull conversation history. If these tools are unavailable, inform the user that this skill only works with AI products that have conversation history features.

Language Configuration

Default: Match the user's language (detect from their message).

If the user specifies a language preference (e.g. "in English", "用中文", "en français"), use that language for the entire report. The HTML report, all analysis text, section headers, and the improvement checklist should all be in the chosen language.

Technical terms, tool names, and conversation titles may remain in their original language regardless of the report language.

Data Collection Phase

Thorough data collection is the foundation of analysis quality. Don't start writing after pulling just one page — the more comprehensive the data, the more valuable the insights.

Step 1: Determine Analysis Window

Default: past 30 days. If the user specifies a different time range, follow their request.

Calculate the time window: today minus 30 days (or the user-specified number of days) to get the after parameter.

Step 2: Pull Conversation Records

Use recent_chats tool multiple times, n=20 each time, paginate with the before parameter until the entire time window is covered or approximately 5 rounds have been pulled.

For each conversation, record:

If the user is inside a Project, only conversations within that Project will be retrieved; if outside a Project, only non-Project conversations are available. Inform the user of this scope limitation.

Step 3: Supplementary Search (Optional)

If high-frequency topics from recent_chats need more context, use conversation_search with relevant keywords to dig deeper.

Analysis Framework

Analyze collected conversation data across five dimensions. Every dimension must be backed by specific evidence (conversation titles or content references) — no abstract conclusions without support.

Dimension 1: Theme Clustering — "What am I using AI for?"

Categorize all conversations by topic. Typical categories include but are not limited to:

Output the conversation count and percentage for each category. Identify the "center of gravity" — the 2-3 areas where time and energy are most concentrated.

Dimension 2: Pattern Detection — "Are there inefficiency patterns?"

Look for these signals:

For each identified pattern, provide specific conversations as evidence.

Dimension 3: Value Output — "Which conversations actually produced something?"

Distinguish two types of conversations:

Estimate the proportion of high-value output conversations. Consumption isn't inherently bad — the goal is to make the ratio visible so the user can judge whether it's healthy.

Dimension 4: Collaboration Friction — "Where does human-AI collaboration break down?"

Look for these friction signals:

For each friction point, analyze root cause: was it a user-side issue (unclear requirements, unrealistic expectations) or an AI-side limitation (capability boundary, tool limitation)?

Dimension 5: Highlights — "When was collaboration at its best?"

Identify 2-3 "best practice" moments:

Analyze what these highlights have in common — under what conditions does collaboration work best?

Output: HTML Report

Generate a polished HTML file, save to /mnt/user-data/outputs/ and present with present_files.

Report Structure

┌──────────────────────────────────────────────┐
│  HEADER                                       │
│  - "AI Usage Audit" + time range              │
│  - Total conversations / coverage / date       │
│  - One-sentence summary (overall profile)      │
└──────────────────────────────────────────────┘
§1  Data Overview
    - Total conversation count, time distribution (which days/weeks most active)
    - Topic distribution bar (pure CSS, no JS chart libraries)
    - Top keywords from the period
§2  Theme Clustering Analysis
    - Each category with conversation count and representative conversations
    - "Center of gravity" analysis
    - 🔍 Insight: Does your AI usage align with your core goals?
§3  Patterns & Efficiency
    - Identified inefficiency patterns (each with specific evidence)
    - Efficiency visualization using "traffic light" notation:
      🟢 Efficient / 🟡 Optimizable / 🔴 Needs change
    - 💡 Improvement suggestions (per pattern)
§4  Value Output Assessment
    - High-value vs. low-efficiency consumption ratio
    - Representative high-value conversations (brief description of output)
    - 📊 ROI assessment: time invested vs. actual output
§5  Collaboration Friction Points
    - Friction type statistics
    - Top 3 friction scenarios with root cause analysis
    - 🔧 Solutions (specific improvement methods per friction point)
§6  Highlights & Best Practices
    - 2-3 highlight moments
    - Common traits behind these highlights
    - ✨ Practices worth replicating
§7  Next Month Improvement Checklist
    - Based on all analysis above, generate 3-5 specific, actionable improvements
    - Each item format: What to do + Why + How to measure
    - Presented in checkbox style for easy tracking
    - This is the most important actionable section — no empty platitudes
§8  One-Paragraph Summary
    - 2-3 sentences for overall closure
    - Tone: like a coach who knows your work giving a monthly debrief
    - Not a repetition of prior sections — distill one core insight

Section Usage Principles

Visual Design

The HTML report follows a clean, modern editorial design:

Overall Style

Key Components

Color System

Google Fonts

Writing Style

Tone

Format

Language

Quality Checklist

Before delivery, verify:

What This Skill is NOT

Credits

Built by Junjie Liu / Philosophie AI

Part of the ClawHub skill collection.

This skill depends on recent_chats and conversation_search tools to pull conversation history. If these tools are unavailable, inform the user that this skill only works with AI products that have conversation history features.