The analyzers surface what to change. These three commands show where your tokens went. All of them respect plan-tier framing, so subscription users see quota shares instead of dollar spend.

`tj cost`

A cost breakdown grouped by agent, model, day, or tool.

tj cost --since 7d --group-by model
tj cost --compare last-week

--group-by accepts agent, model, day, or tool. --compare takes a prior window (previous, last-week, last-month, last-7d, last-30d, or a date range) and renders a window-over-window diff with ▲/▼ indicators, per-agent and per-model top shifts, and dollar plus token deltas. On subscription plans the cost cells render as a share of the cycle rather than raw dollars.

`tj tokenmaxx`

A screenshottable quota/efficiency card.

tj tokenmaxx
tj tokenmaxx --weekly

It leads with context composition: what share of your quota went to overhead (re-reading history, CLAUDE.md, accumulated tool output) versus real work. That share picks an efficiency tier, from TokenMinimizer through QuotaSink, with lower overhead being the leaner tier. --weekly is a 7-day recap preset. The efficiency number is a measured token share, never a guaranteed saving, and dollars appear only as a secondary line for API plans.

`tj context`

The context-cost diagnostic over your Claude Code sessions.

tj context --since 30d
tj context --agent my-agent

It renders per-turn context composition (re-reading versus net-new work), recurring inclusions (the same file re-read across many sessions, each with a concrete @file or CLAUDE.md fix), and compact candidates (sessions where a /compact reclaims the most quota). Headline numbers are framed in quota terms for subscription users; dollars are a secondary signal for API users.

Cost visibility

`tj cost`

`tj tokenmaxx`

`tj context`

See also