ls-insights-deep

name	ls-insights-deep
description	Deep usage insights across ALL Claude Code sessions (full history, not just recent). Scans every JSONL conversation file, merges with facets and session-meta, then generates a beautiful visual HTML report using the mockup skill. Use when you want comprehensive analysis covering months of sessions, not just the last 50-60.

Deep Insights

Full-history Claude Code usage analysis. Covers all sessions back to the beginning, not just recent ones.

Step 1: Build the Data

Run the aggregation script to scan all sessions:

python "${CLAUDE_SKILL_DIR}/build_insights_data.py"

This outputs ~/.claude/usage-data/insights_data.json with all sessions merged.

If it takes more than 2 minutes, show a progress note to the user.

Step 2: Load and Summarize

Read ~/.claude/usage-data/insights_data.json.

Extract for analysis:

• summary block (totals, date range, top projects, top tools)
• All sessions where source == "facet" (richest data — include ALL of these)
• A representative sample of session_meta sessions (up to 150)
• First-prompt data from raw JSONL sessions to understand topic spread

Step 3: Parallel Analysis

Spawn 4 subagents in parallel, each focused on a different lens. Give each subagent the full facets array plus the summary stats. Ask each to return structured JSON.

Agent A — Project & Topic Landscape

• What projects dominate (by session count, commits, tokens)?
• What are the top 5-8 distinct work areas?
• How has focus shifted over the date range (early vs. recent periods)?
• Which projects got the most debugging effort vs. new feature work?

Agent B — Productivity & Velocity Patterns

• Sessions per week/month trend
• Commits per session, lines added/removed trends
• Tool usage patterns (Bash vs. Read vs. Edit ratios, MCP usage, web fetch)
• Session length distribution (short <5 msg vs. marathon >50 msg)
• Peak productivity hours/days from message_hours data

Agent C — Friction & Quality Analysis

• Aggregate all friction_counts across all facet sessions
• Top friction categories ranked
• Sessions by outcome (fully_achieved vs partial vs not_achieved)
• Claude helpfulness distribution
• Common tool errors and error categories
• Sessions where user_interruptions were high

Agent D — Evolution & Growth

• Compare earliest 25% of sessions vs. most recent 25%
• Has friction decreased over time?
• New tools/capabilities adopted over time
• Goal complexity trend (single_task vs multi_task vs complex)
• Most ambitious sessions (highest tool counts, most files modified)

Step 4: Synthesize

After all 4 agents complete, synthesize into a single insights structure:

{
  "headline_stats": { ... },
  "project_landscape": { ... },
  "productivity_patterns": { ... },
  "friction_analysis": { ... },
  "evolution": { ... },
  "top_sessions": [ ... ],
  "recommendations": [ ... ]
}

Step 5: Generate the Report

Load the mockup skill and generate a beautiful self-contained HTML report:

Skill tool → skill: "mockup"

The report should use the Data Visualization Mode from the mockup skill.

Save to: ~/.claude/usage-data/insights_deep_report.html

Then open it:

start "~/.claude/usage-data/insights_deep_report.html"

Report it done with the file path.