usage-limit-reducer audits Claude Code token waste

This is a practical observability layer for Claude Code, not a magic saver: it turns vague limit anxiety into a concrete token report and a short list of fixes.

• –The repo reads `~/.claude/projects/*/*.jsonl` locally, so the analysis stays on-device instead of sending session data elsewhere.
• –Its strongest idea is prioritization: it applies only the 2-4 rules that match the current session, which should keep advice actionable.
• –The project’s own example highlights history re-reading as the dominant token sink, which is the right problem to target in long chats.
• –The suggested next actions map cleanly to real operator behavior: `/compact`, `/clear`, `/model claude-haiku-4-5`, or adding `CLAUDE.md` for reusable context.
• –Because this is a skill, not a dashboard, the value is in changing how people work inside Claude Code rather than just reporting stats.