Claude Code Exposes Hidden Token Drains
A tutorial explains how Claude Code’s `/usage` diagnostics expose token consumption from background loops, cross-session messages, goal check-ins, agent teams, long contexts, and cache misses. It also shows settings developers can use to limit unnecessary spend.
Claude Code’s biggest cost problem is often invisible orchestration, not the code it writes. The new diagnostics make agent economics measurable—and expose why “idle” sessions can still consume meaningful usage.
- –`/usage` breaks usage down by models, loops, subagents, skills, plugins, and MCP servers.
- –Scheduled tasks, cross-session messages, goal check-ins, and prompt suggestions can trigger requests while a session appears idle.
- –Cache misses after long breaks reprocess the full context, creating sharp usage spikes.
- –Agent teams multiply costs because every teammate maintains its own context window; small teams and focused prompts matter.
- –Developers can reduce waste by clearing stale sessions, disabling unnecessary check-ins, holding inbound messages, and choosing cheaper models for coordination.
DISCOVERED
1h ago
2026-10-12
PUBLISHED
1h ago
2026-10-12
RELEVANCE
AUTHOR
DIY Smart Code