# Check Usage > Check Claude Code's current 5-hour rolling session usage and weekly usage percentages against your subscription limits. Use when the user asks about usage limits, how much of their session/quota they've used, whether they're close to being rate limited, or when to /compact to save context. - Skill: `mhollander38/check-usage` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add mhollander38/check-usage` - Raw SKILL.md: https://api.skillmd.com/api/skills/mhollander38/check-usage/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: mhollander38 (https://skillmd.com/u/mhollander38) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/mhollander38/check-usage --- # Check Usage Run the bundled script to get live usage numbers from your Anthropic account: ``` bash /scripts/check-usage.sh ``` (Use the base directory shown when this skill was invoked to build the path.) The script prints lines like: ``` Current session: 77% used · resets Sep 7 at 9:29pm (Europe/London) Current week (all models): 50% used · resets Sep 13 at 4:59am (Europe/London) ``` Report both percentages and their reset times back to the user in a friendly, concise way. Do not fetch or relay the rest of the `/usage` breakdown (subagent/skill/plugin stats) — only the session and week lines are relevant here. ### Weekly exhausted but session shows room — always flag this A healthy session percentage does not mean capacity is fine on its own. If the script prints a `WARNING:` line, that means weekly is at/near 100% while the session limit still shows room. **Do not fold this into a routine "here's your usage" report — call it out explicitly and first**, before the plain percentages: say clearly that weekly quota is exhausted, and that continuing to run right now either bills purchased usage credits (real money, if enabled) or is running on borrowed time before a hard stop. Point them at `/usage-credits` (interactive-only) to confirm which. This applies even when following the Usage Capacity Policy's "keep going, don't stop to ask" default — keep going, but never silently. ### Extra / purchased usage ("usage credits") Anthropic's pay-as-you-go feature for continuing past the plan's included quota is called **usage credits** in the current CLI (older internal naming: "overage"/"extra usage" — you may still see those words in messages). It's off by default and is enabled per-account or by an org admin, so most runs of this script won't see anything about it. If the script's output *does* include a line mentioning credits/extra usage/overage — e.g. a credit balance, spend amount, or "usage credits are off" — report that alongside the session/week percentages, in plain terms (e.g. "you also have $X of purchased usage credits available/spent"). There is no separate flag or JSON field for this in Claude Code 2.1.269 — it only ever surfaces as plain text inside `/usage`'s output, so this script's regex is the only hook into it. If the user asks to enable, buy, or check their usage-credit balance/settings, tell them to run `/usage-credits` in an **interactive** Claude Code session — it opens `claude.ai/settings/usage` in a browser and can't be done headlessly. On a Team/Enterprise plan, an org admin manages it instead, at `claude.ai/admin-settings/usage`. If the script's output doesn't match the expected format (e.g. the account is on API-key billing rather than a subscription), just relay whatever it printed and note that structured percentages aren't available.