What problem does it solve? Claude Code users have no built-in way to see how many tokens they consume per project, model, or session, making it hard to track spend, enforce model-selection policies, or avoid hitting five-hour billing block limits during long loops. ## Core Features & Use Cases - Usage reporting via ccusage: Runs pinned ccusage commands (daily, weekly, monthly, session, blocks) against local ~/.claude/projects/**/*.jsonl transcripts and summarizes totals, top projects, and per-model spend. - Agents mode: Queries the agents-observe plugin REST API for per-subagent token attribution, delegation chains, and model-policy compliance within a specific session. - Loop circuit breaker: Checks the active five-hour billing block, projects burn rate, and emits a PASS/STOP verdict for /loop safeguards. - Use Case: Before starting a long automated loop, run the blocks mode to confirm the current five-hour block has enough headroom, then review weekly per-model spend to catch Opus usage on read-only exploration tasks. ## Quick Start Ask the assistant to run /usage-report to show this week's token usage and cost breakdown by project and model.
