~/.claude — session-report
> claude usage
loading…
—

findings

98.4%
of the week's tokens ran under the old Desktop/AI-Workspace path — only 1.6% under the new AI-Workspace location, so the migration hasn't shifted real usage yet
5.8%
of the week went to a single bare "1" reply — 152 API calls and 9 subagent calls continuing an already-huge context
47.8%
of the week's tokens landed on one day, 2026-07-12, across a burst of long-running sessions
97.8%
overall cache-hit rate — context reuse is healthy across the week
13×
cache breaks over 100k tokens this week, clustering around 07-12 to 07-15 — repeated expensive re-warms, mostly at session handoffs

summary

tokens by projectshare of total

session timeline by dayclick a day · ←/→ to navigate

—
00:0006:0012:0018:0024:00

most expensive promptsclick to expand context

cache breaks>100k uncached · click for context

projects

subagent types

skills & slash commands

recommendations

Almost all work still ran under ~/Desktop/AI-Workspace (98.4% of tokens, 35/40 sessions) even though the repo has moved to ~/AI-Workspace — worth confirming new terminals/sessions are actually opened in the new path, or the split-state migration risk this workspace's own CLAUDE.md warns about will keep compounding.
The "1" prompt that cost 31.8M tokens (152 calls) suggests a very long-running session got a one-character continuation instead of a fresh context — consider starting a new session (or /clear) after a big milestone rather than replying tersely into an already-large transcript.
Cache breaks cluster right at session-handoff prompts ("重新交接", "继续往下做吧") — if handoffs across terminals are routine, writing the continuation prompt to platform/prompts/ (as this workspace's CLAUDE.md already asks) before closing a session avoids re-priming a multi-hundred-k-token context cold.