Monitor your Claude credit pool.
Align your window.

Since June 2026, claude -p / Agent SDK / GitHub Actions draw from a separate monthly credit pool. This watches that pool and reminds you to start your session — on your always-on box.

credit-pool monitor window reminder ccusage-powered no model calls

Get started View on GitHub →

What it does

Two hands-off jobs. It only reports and reminds — it never automates the model and never circumvents a limit. Usage is read locally with ccusage (no model calls, no window side effects).

① Monitor the credit pool

Month-to-date burn, projected month-end spend, and an alert at your thresholds — so automations don't quietly stop when the pool (no rollover) runs dry.

② Remind you

At your anchor time it nudges you to send your first interactive prompt, so your 5-hour window lines up with your working day. You stay the one doing interactive usage.

programmatic credit pool — spent this month$0.00

projected by month end$0.00

cap $100 (Max 5x) · ▏ warn 80% · ▏ critical 95% — alert fires before you run dry

check → report → remind:

Setup — 1, 2, 3

Run on your always-on box (Node 18+; npm i -g ccusage for the fast path).

npx skills add DecisionVex/claude-usage-companion   # skills.sh
clawhub install claude-usage-companion              # OpenClaw / ClawHub

1 Configure

cp config.example.json config.json
# set "timezone", "plan" (pro|max5x|max20x), "anchor"

2 Check & see your burn

node bin/usage-companion.mjs check
node bin/usage-companion.mjs report

3 Schedule

node bin/usage-companion.mjs install   # prints a cron block
crontab -e                              # paste it, save

Configuration

Field	Meaning	Default
`timezone`	IANA tz for all clock math — set this	`UTC`
`plan`	`pro`/`max5x`/`max20x` → credit cap ($20/$100/$200)	`max5x`
`monthly_credit_usd`	Override the cap	`null`
`warn_pct` / `critical_pct`	Alert thresholds (% of cap)	`80` / `95`
`anchor`	Local time you start work	`08:00`
`alert_command`	Shell command receiving alerts on stdin (e.g. Slack). `null` = stdout + log	`null`

Why this exists

In June 2026 Anthropic moved non-interactive usage — Agent SDK, claude -p, Claude Code GitHub Actions, third-party agents — onto a separate monthly credit pool, metered at API rates with no rollover. Interactive limits stay reserved for hands-on use.

✓ What this does — observability + a human reminder, fully within terms: reads your own usage locally, alerts you before the credit pool empties, and nudges you to start your interactive session.

✗ What it never does:

automate the model or script interactive input to "warm" the window
extract or reuse credentials
circumvent or exceed any limit

Successor to claude-session-warmer, whose claude -p priming no longer anchors the window after this change.