Feature: Incremental /insights with checkpoint support

Problem

/insights re-analyzes all sessions from scratch every time it runs. As session count grows, this becomes slower and redundant — most sessions have already been analyzed in a previous run.

Proposed Solution

Add checkpoint support so /insights can pick up where it left off:

1. After each /insights run, save a checkpoint (e.g., timestamp or session ID of the last analyzed session) to ~/.claude/usage-data/checkpoint.json
2. On subsequent runs, only analyze sessions created after the checkpoint
3. Merge new analysis with previous results to produce the full report

Optional enhancements

• --since <date> flag to manually specify a start date
• --full flag to force a full re-analysis ignoring the checkpoint
• Trend tracking between checkpoints (e.g., "friction went down this week", "you shipped 3x more commits than last period")
• Diff view showing what changed since last insights run

Why This Matters

Power users who run /insights regularly (weekly, biweekly) would benefit from:

• Faster execution (only processing new sessions)
• Trend analysis over time (comparing checkpoint periods)
• Less redundant LLM token usage on already-analyzed sessions

🤖 Generated with Claude Code