insights-report - SKILL.md Agent Skill

name: insights-report description: Use when the user wants to review their Claude Code usage insights, see session statistics, understand interaction patterns, or get a summary of how they use Claude Code

Insights Report

Generate, open, and extract actionable recommendations from Claude Code usage insights.

Workflow

digraph insights_flow {
    "Start" [shape=doublecircle];
    "Run /insights" [shape=box];
    "Report exists?" [shape=diamond];
    "Open in browser" [shape=box];
    "Read & extract" [shape=box];
    "Done" [shape=doublecircle];

    "Start" -> "Run /insights";
    "Run /insights" -> "Report exists?";
    "Report exists?" -> "Open in browser" [label="yes"];
    "Report exists?" -> "Tell user to run /insights manually" [label="no"];
    "Open in browser" -> "Read & extract";
    "Read & extract" -> "Done";
}

Step 1: Generate Insights

Run the /insights command using the Skill tool:

Skill tool: skill: "insights"

If the Skill tool fails (built-in command), use Bash:

claude -p "/insights" --allowedTools "Read,Bash,Glob,Grep"

If both fail, ask the user to run /insights manually, then proceed to Step 2.

Step 2: Open Report in Browser

The report lives at ~/.claude/usage-data/report.html.

Open it in the default browser:

start "" "C:/Users/simon/.claude/usage-data/report.html"

Tell the user the direct URL: file:///C:/Users/simon/.claude/usage-data/report.html

Step 3: Extract Actionable Content

Read the report HTML from ~/.claude/usage-data/report.html.

Do NOT output statistics (message counts, session counts, tool usage numbers, charts, etc.).

Instead, output the following sections in this exact order:

1. Where Things Go Wrong

Extract from the section-friction area. Output as bullet points:

For each friction category, output the title as a bold bullet
Under it, output the description as a sub-bullet
Under that, output each example as a sub-sub-bullet

2. Existing CC Features to Try

Extract from the section-features area. Output ALL content:

CLAUDE.md Additions: For each suggested CLAUDE.md item, output:

The full text of the CLAUDE.md addition (from the data-text attribute or cmd-code element) as a fenced code block
The "why" explanation underneath

Feature Cards: For each feature card (Custom Skills, Hooks, Headless Mode, etc.), output:

The feature title and one-liner description
The "Why for you" explanation
The full example code block exactly as shown

3. New Ways to Use Claude Code

Extract from the section-patterns area. Output ALL content:

For each pattern card, output:

The pattern title and summary
The full detail/explanation text
The full "Paste into Claude Code" prompt as a fenced code block

4. Headless Mode Command Lines

Collect all headless mode (claude -p ...) command lines from anywhere in the report. Output each as a fenced code block with a brief label of what it does.

5. On the Horizon

Extract from the section-horizon area. Output ALL content:

For each horizon card, output:

The title
The full "what's possible" description
The "Getting started" tip
The full "Paste into Claude Code" prompt as a fenced code block

6. Output

Output a markdown file ./insights_report_{timestamp}.md