usage-checker

name	usage-checker
description	This skill should be used when the user asks "check my usage", "how's my Claude usage", "am I running low", "which account has capacity", "check usage", "how much usage do I have left", "usage report", "burn rate", "how fast am I burning", "which account should I use", mentions Claude Max rate limits, or asks about any of their Claude Max accounts.

Claude Max Usage Checker

Check usage across all configured Claude Max accounts and report results in plain language.

Checking Usage

Run this command to get structured JSON data for all accounts:

~/code/claude-usage/claude-usage check --json 2>/dev/null

Parse the JSON output and present a human-readable summary. Do not show raw JSON to the user. The number of accounts is determined by the output — do not assume a fixed count.

To check a single account by number or name:

~/code/claude-usage/claude-usage check --json -a pelotom 2>/dev/null

Interpreting Results

For each account, extract these fields from the JSON api_data:

• 5-hour session usage — five_hour.utilization (0-100%). The short-term rate limit.
• 7-day weekly usage — seven_day.utilization (0-100%). The main weekly limit. This is the most important number.
• Sonnet-only usage — seven_day_sonnet.utilization. Separate limit just for Sonnet.
• Reset times — resets_at fields. Convert to human-relative times ("resets in 4h", "resets tomorrow").
• Subscription status — from subscription_details data: next_charge_date, status.

Pace Assessment

Compare the percentage of weekly usage consumed against the percentage of the 7-day window that has elapsed:

• time_remaining = resets_at - now
• window_elapsed = (7 days - time_remaining) / 7 days
• pace = utilization / (window_elapsed * 100)

Describe pace in plain language:

• pace < 0.5 → "plenty of room"
• pace 0.5–0.8 → "comfortable"
• pace 0.8–1.0 → "on pace to use the full limit"
• pace 1.0–1.5 → "running a bit hot"
• pace > 1.5 → "burning through usage fast"

If resets_at is null, no usage has been recorded — report as "completely fresh."

Choosing an Account

When the user asks which account to use, recommend the one with the lowest 7-day utilization. Factor in reset times — an account at 60% that resets in 2 hours is better than one at 30% that resets in 6 days.

Handling Errors

If a result has "error" set (non-null), the session has expired. Instruct the user:

Run: ~/code/claude-usage/claude-usage setup -a {account_number}

This opens a browser for Google OAuth re-login. The user must run this in their own terminal — it cannot be run from within Claude Code.

Account Management

Accounts are configured in ~/code/claude-usage/accounts.json. To see which accounts exist:

cat ~/code/claude-usage/accounts.json

First-time setup or session refresh

~/code/claude-usage/claude-usage setup          # all accounts
~/code/claude-usage/claude-usage setup -a 2     # just account #2

Setup opens a Playwright browser for each account. The user logs in via Google OAuth. Cookies are exported to ~/code/claude-usage/profiles/account-{N}/cookies.json. Sessions typically last weeks before expiring.

Example Response Style

When the user asks "how's my usage?", respond conversationally:

pelotom is at 57% of the weekly limit with 22 hours until reset — comfortable pace. thomasmcrockett has barely been touched (3% weekly). eumemic is completely fresh. I'd use eumemic or thomasmcrockett next.

Lead with the most important information (which accounts are getting close to limits). Only mention billing dates if the user asks.

Burn Rate Analysis

For trend analysis over time, use the report command:

~/code/claude-usage/claude-usage report 2>/dev/null
~/code/claude-usage/claude-usage report --json 2>/dev/null
~/code/claude-usage/claude-usage report --hours 6 -a eumemic 2>/dev/null

This reads from JSONL history (~/code/claude-usage/data/usage-history.jsonl) populated by the track command. It computes burn rates via linear regression and projects ETAs to limit.

If the user asks about burn rate and there's no history data, tell them to set up tracking:

# Record snapshots (run via cron every 15 minutes):
~/code/claude-usage/claude-usage track

# Cron entry:
# */15 * * * * ~/code/claude-usage/claude-usage track 2>>~/code/claude-usage/data/track-errors.log

When presenting report data, focus on:

• Which accounts are burning fastest (highest slope_per_hour)
• Which will hit the limit before reset (will_hit_limit flag)
• The recommendation for which account to use next

Pool Health Analysis

The report command includes pool-level analysis that accounts for token rotation.

Key Concepts

• Pool headroom: Total remaining capacity across all uncapped accounts. E.g., if 4 accounts are at 20%, 50%, 80%, 100%, headroom = 80+50+20 = 150%.
• Active burn rate: The rate of the currently-burning account (highest slope among uncapped). Only one account burns at a time due to rotation.
• Pool runway: headroom / active_burn_rate. How long until ALL accounts are capped if burning continues at the current rate with sequential rotation.
• Safe if reset comes first: If any capped account resets before the pool runway expires, capacity is replenished.

Alert Levels

• ok: Multiple accounts have headroom, runway is comfortable
• warning: Down to 1-2 accounts, or both remaining are above 80%
• critical: All accounts capped, or last account hits cap in <2h

Example Interpretation

Pool headroom: 150% across 3 accounts Active burn: +2.5/hr (sequential rotation) Runway: ~60h until all accounts capped Next reset: 18h

This means: at 2.5%/hr, 150% of headroom lasts 60 hours. The first account resets in 18 hours, well before exhaustion. The pool is healthy.

Additional Resources

• references/api-format.md — Full JSON response schema, field definitions, and pace calculation details