Show HN: CostClaw, a local CLI that finds your Claude Code spend leaks

{“telegram”:”CostClaw audits Claude Code logs locally and prices the waste\n\nA Show HN project called CostClaw is a console tool that parses Claude Code session logs in JSONL straight off disk, including the sublogs from subagents and workflows. Parsing happens in memory and nothing leaves the machine; you run it with npx costclaw audit.\n\nThe report hunts four kinds of waste: cache misses (cache reads run $0.30 per million tokens against $3.00 for input), sessions that hammer a single tool a hundred times, short tasks sent to Opus at $15 per million tokens instead of Sonnet’s $3, and long sessions where caching stops paying for itself. It also scores six areas separately, from CLAUDE.md quality to spending discipline.\n\nEvery build seeds the test logs with traps: a fake prompt and a client path. The CLI is free, and costclaw optimize unlocks with a one-time $39 license.”}
Up next
The same three Claude models are throwing errors again
Anthropic says it has identified the cause of elevated error rates on Claude Mythos 5, Fable 5 and Opus 5, flagged at 07:05 UTC on August 5, with a fix in progress. Same trio broke July 25. claude.ai, the API, Claude Code and Cowork all affected.
Related stories
- Codedict cuts Claude Code tokens 63%, but cost only 26%
- AI Usage Ball shows Claude quota as liquid orbs on macOS
- ctx2img packs whole repos into image tiles, cutting token use 60-75%
- Intent-lens spells out Claude Code calls before they run
- claude-files-pane shows Claude Code’s file reads live
- lcc cut one Claude Code session 49.5% without a summary
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

