claude-code-metrics-analysislisted

# Claude Code Metrics Analysis ## When to Use - Periodically (weekly/monthly) to spot UX friction trends - After noticing slowness, rate limits, or unexpected costs - When deciding which permissions to auto-approve or which workflows to optimize - When investigating whether Claude Code usage patterns are efficient ## Data Sources All data lives under `~/.claude/`: | Source | Path | Format | Contains | |--------|------|--------|----------| | Stats cache | `stats-cache.json` | JSON | Aggregate usage: daily activity, model tokens, hour distribution, session totals | | Telemetry | `telemetry/*.json` | JSON lines | Granular events: tool use, errors, stalls, rate limits, session lifecycle, performance | | History | `history.jsonl` | JSONL | User prompts with timestamps, paste events, session IDs, project paths | ## Analysis Procedure ### Step 1: Load aggregate stats Read `~/.claude/stats-cache.json` and extract: - **Usage trends**: daily message/session/tool counts — look for spikes or drops - **Hour distribution** (`hourCounts`): identify peak hours and potential burnout patterns - **Model usage**: token consumption by model, cache hit ratios - **Session extremes**: longest session, avg messages/session ### Step 2: Extract telemetry event counts Run a frequency count across all `~/.claude/telemetry/*.json` files: ```bash cat ~/.claude/telemetry/*.json | python3 -c " import sys, json, collections events = collections.Counter() for line in sys.stdin: try: d =