How do I install the Check Usage skill?

Run npx skillmds@latest add jgjake2/check-usage in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Check Usage safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Check Usage?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Check Usage free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Check Usage?

jgjake2 (@jgjake2) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Check Usage by jgjake2 — Agent Skill

Name: Check Usage
Author: jgjake2

Check Usage

Check Claude's current rate-limit utilization and decide whether to pause the task. Use before starting an unusually expensive step of a long-running task, when the usage-limit hook is not installed, or when the user asks where the budget stands.

jgjake2 27b24f4 2 files · 3.6 KB Updated Sep 17, 2026

File contents

check-usage

node .claude/skills/check-usage/check-usage.js

Prints a single JSON object to stdout.

If the PreToolUse usage-limit hook is installed in this project, checking is already automatic (every 10 minutes, after a task has run 30 minutes). Do not poll this on a timer as well -- each check makes a real API call upstream.

Response

Field	Meaning
`band`	`ok`, `warn`, `grace`, `stop`, or `unknown`
`shouldStop`	`true` only for `stop`
`degraded`	`true` when the reading is unavailable OR only partly understood -- unknown, not safe
`reason`	One-line summary including any caveats
`windows`	Every limit window reported, each with `utilization` (0-1), `percent`, `resetsAt`
`highest`	The window closest to its limit -- this drives the verdict
`grace`	When banded `grace`: which window, when it resets, how long until then
`unreadableWindows`	Windows whose utilization could not be parsed; non-empty means the verdict is partial
`ignoredWindows`	Windows skipped because their `status` marks them inapplicable
`configIssues`	Threshold/env problems; non-empty means defaults were substituted

How to act on it

ok -- continue normally.
warn (>=80%) -- finish the current step; do not begin a large new one. Consider a checkpoint via save-progress. Advisory only.
grace -- past the stop threshold, but the limiting window resets within 10 minutes. Keep working. Do not halt and do not checkpoint for this. Avoid starting a large step until after grace.resetsAt.
stop (>=90% with no imminent reset, or >=98% regardless) -- halt. Write up the work so far and a concrete resume plan, save both with save-progress, then stop and tell the user how to resume.
unknown / degraded: true -- keep working. Never treat a missing reading as a reason to stop, and never assume usage is low. Mention it to the user once.

Notes

Exit code is always 0 when the check ran. A non-zero exit means the script itself broke.
All limit windows are read generically, so a separate Opus weekly cap is picked up without a code change.
Thresholds are set as fractions (0.9, not 90) via USAGE_LIMIT_STOP_AT, USAGE_LIMIT_WARN_AT, USAGE_LIMIT_HARD_STOP_AT, and USAGE_LIMIT_RESET_GRACE_MS. An out-of-range value falls back to the default and is reported in configIssues rather than silently disabling the check.

---name: check-usagedescription: Check Claude's current rate-limit utilization and decide whether to pause the task. Use before starting an unusually expensive step of a long-running task, when the usage-limit hook is not installed, or when the user asks where the budget stands.---# check-usage```node .claude/skills/check-usage/check-usage.js```Prints a single JSON object to stdout.**If the PreToolUse usage-limit hook is installed in this project, checking is already automatic**(every 10 minutes, after a task has run 30 minutes). Do not poll this on a timer as well --each check makes a real API call upstream.## Response| Field | Meaning ||---|---|| `band` | `ok`, `warn`, `grace`, `stop`, or `unknown` || `shouldStop` | `true` only for `stop` || `degraded` | `true` when the reading is unavailable OR only partly understood -- unknown, not safe || `reason` | One-line summary including any caveats || `windows` | Every limit window reported, each with `utilization` (0-1), `percent`, `resetsAt` || `highest` | The window closest to its limit -- this drives the verdict || `grace` | When banded `grace`: which window, when it resets, how long until then || `unreadableWindows` | Windows whose utilization could not be parsed; non-empty means the verdict is partial || `ignoredWindows` | Windows skipped because their `status` marks them inapplicable || `configIssues` | Threshold/env problems; non-empty means defaults were substituted |## How to act on it- **`ok`** -- continue normally.- **`warn`** (>=80%) -- finish the current step; do not *begin* a large new one. Consider a checkpoint via `save-progress`. Advisory only.- **`grace`** -- past the stop threshold, but the limiting window resets within 10 minutes. **Keep working.** Do not halt and do not checkpoint for this. Avoid starting a large step until after `grace.resetsAt`.- **`stop`** (>=90% with no imminent reset, or >=98% regardless) -- halt. Write up the work so far and a concrete resume plan, save both with `save-progress`, then stop and tell the user how to resume.- **`unknown`** / `degraded: true` -- **keep working**. Never treat a missing reading as a reason to stop, and never assume usage is low. Mention it to the user once.## Notes- Exit code is always 0 when the check ran. A non-zero exit means the script itself broke.- All limit windows are read generically, so a separate Opus weekly cap is picked up without a code change.- Thresholds are set as **fractions** (`0.9`, not `90`) via `USAGE_LIMIT_STOP_AT`, `USAGE_LIMIT_WARN_AT`, `USAGE_LIMIT_HARD_STOP_AT`, and `USAGE_LIMIT_RESET_GRACE_MS`. An out-of-range value falls back to the default and is reported in `configIssues` rather than silently disabling the check.

Frequently asked questions