Cchubber

CC Hubber

Your Claude Code usage, diagnosed. One command.

Reads your local data, generates an HTML report. No API keys, no accounts. Sends anonymous stats for community benchmarks (details). Opt out: --no-telemetry.

What you get

A single HTML report that tells you three things: what you spent, why you spent it, and whether that's normal.

The diagnosis:
- Cache health grade (trend-weighted, recent 7 days count more)
- Inflection point detection: "Your efficiency dropped 3.2x starting March 17"
- Per-project cost breakdown with decoded project names
- Session intelligence: duration stats, tool usage, activity heatmap
- Model routing analysis (93% Opus? Your limits would last 3x longer on Sonnet)
- 8 actionable recommendations, each with estimated usage savings

The data:
- Cost calculated from actual token counts (LiteLLM pricing, not the broken costUSD field)
- Message-level deduplication (Claude Code JSONL files contain ~50% duplicates from session resume)
- Subagent visibility: Haiku and Sonnet background agents show up in model distribution
- CLAUDE.md section-by-section analysis with per-message cost impact
- Cache break estimation even when diff files don't exist on your CC version

Your week against your plan:
Directly under the "Did you spend them well?" strip, the report shows how much of your last 7 days of Claude Code (and Codex) work went to what you said you'd do: one number, what it was measured against, where the rest went, and the day you drifted most. "What you said you'd do" is read from a PLAN.md, plan.md or TODO.md checklist in the project you work in (or a Mover OS Daily Note). With no written plan, or no work in the window, the section says what to add instead of showing a number. It runs on your machine and nothing is uploaded; detour names…