1. Home
2. Journal
3. Measure your work with Claude Code

Claude Code · Codex

Measure your work with Claude Code: /usage, /insights and ccusage

Three tools to see how much you use your agents and how you work with them. What each one shows, and what none of them keeps.

Published September 29, 2026 · 5 min read

How many sessions this month? Which model? Are your prompts getting longer? Claude Code answers some of these questions, and a community tool covers the rest. They all share one limit: they read your Mac's local history.

/usage: how much, lately

Inside a session, /usage starts with the current session's details: tokens per model, duration and a cost estimate at list price. On a Pro or Max subscription, that cost isn't used for billing; what you mostly see is your plan's usage bars and activity stats.

On Pro, Max, Team and Enterprise plans, /usage also breaks down what counts against your limits: the share of skills, subagents, plugins and each MCP server, costly behaviors (long context, cache misses) and your heaviest scheduled tasks. The d and w keys switch between the last 24 hours and the last 7 days.

These numbers are approximate and computed from the session history on this Mac: what you do on another computer or on claude.ai isn't included.

/insights: how you work

/insights doesn't count tokens: it reads your recent sessions and writes an HTML report on how you work. What you work on, friction points (misunderstood requests, buggy code) and suggestions for using Claude Code better.

• Each run analyzes up to 200 sessions it hasn't seen yet, and skips very short ones.
• The report is written to ~/.claude/usage-data/report.html, with a timestamped copy for every run.
• The analysis runs through your usual account: it spends tokens from your plan or API.

ccusage: the community counter

ccusage is a command-line tool, independent of Anthropic and OpenAI, that reads your agents' local JSONL files and turns them into usage reports. Nothing to install:

npx ccusage@latest daily

It offers daily, weekly, monthly and per-session reports (daily, weekly, monthly, session). It reads Claude Code and Codex, among others: ccusage claude daily or ccusage codex daily to target one of them.

What none of the three does

These tools answer "how much." They don't keep what you asked, or what came out of it. And since they read local history, their numbers stop where Claude Code cleaned up: a quarter later, the early sessions are gone unless you changed the default.

That's where Opuscule's activity report comes from. Every volume includes a report on your quarter: sessions per month, prompt length, agent actions per prompt, models used, records. And the rest of the book shows the sessions themselves.