🌅 goodmorning claude

"gooodmorning claudeee!!!" — your Claude Max/Pro 5-hour window starts while you're still asleep.

Landing page: https://marcozambrella.github.io/goodmorning-claude/

The Problem

Claude Max gives you a generous usage window that resets every 5 hours, but the timer only starts at your first message. Start working at 10:00 AM, hit your limit at noon — you're now stuck until 3:00 PM, waiting the full 5 hours.

The Solution

A tiny automation on GitHub Actions (works even when your computer is off) that sends Claude a "gooodmorning claudeee!!!" the moment your previous window expires. Example:

• 🌅 Automation sends the greeting at 8:00 AM → 5-hour window starts immediately.
• 💻 You sit down to work at 10:00 AM.
• ⏳ If you hit your limit at noon, reset lands at 1:00 PM: wait 1 hour instead of 3.

The schedule is dynamic: the script stays in sync with your real reset time — the one Anthropic's API reports — and always restarts from your last session. If a window is already active, it does nothing at all.

How It Works

The workflow (.github/workflows/good-morning.yml) keeps a single run alive for ~6 hours and calls automation/good_morning.py every 5 minutes from inside that run — so the cadence never depends on GitHub's (unreliable) scheduler. Each call:

1. Reads the real reset time of your 5-hour window from automation/state.json (saved by the previous run).
2. Window still active? → exits immediately. Zero API calls, zero quota used.
3. Window expired (or state unknown)? → sends a minimal 1-token "gooodmorning claudeee!!!" directly to the Anthropic API (Haiku model). That single call does double duty: it starts the new 5-hour window and returns the anthropic-ratelimit-unified-5h-reset response header — the true reset time of your account.
4. Saves that reset time back to state.json (committed by the workflow). The next runs know exactly when the window expires — even if you started it yourself by using Claude before the automation fired.

Setup (5 minutes)

1. Generate your OAuth token (requires Pro/Max plan)

On your computer where you're already logged into Claude Code:

claude setup-token

Copy the printed token — it's valid for 1 year.

2. Set the secret in your repo

gh secret set CLAUDE_CODE_OAUTH_TOKEN --repo <your-user>/goodmorning-claude

(paste the token when prompted), or via GitHub: Settings → Secrets and variables → Actions → New repository secret.

3. Done!

The workflow runs automatically. Test it instantly from Actions → goodmorning claude → Run workflow.

Customization

In automation/good_morning.py:

Constant	Default	Description
GREETING	"gooodmorning claudeee!!!"	Message sent to Claude
CLAUDE_MODEL	claude-haiku-4-5-20251001	Model used (Haiku = minimum cost against quota)
WINDOW_HOURS	5	Fallback estimate, used only if the reset header ever disappears

You can also narrow the schedule in the workflow (e.g. 2-59/5 5-23 * * *). Because ticks are what re-arm the loop, restricting them to daytime effectively pauses coverage overnight — handy if you don't want windows launched while you sleep.

Notes

• The script reads the window state from the official anthropic-ratelimit-unified-5h-* response headers of the API call itself — the same numbers Claude Code sees. (The dedicated usage endpoint can't be used: tokens from claude setup-token lack the user:profile scope it requires.)
• The pre-warm message costs a negligible fraction of quota: a single 1-token Haiku turn, and only when a new window actually needs starting.
• Even if you hit your limit (HTTP 429), the response headers still carry the reset time, so the script stays in sync instead of failing.
• Reliable on GitHub alone — no external service, no extra secret. GitHub's scheduled crons are best-effort: under load they are delayed or skipped, sometimes for hours. So coverage does not rely on them firing on time. Instead, one run loops internally every 5 minutes for ~6 hours (the runner's hard limit), restarting the window the instant it expires. When that run ends, a scheduled tick that piled up in the queue starts the next run immediately — concurrency keeps exactly one tick waiting in the wings, so the baton is always there to pass. The cron: 2-59/5 is just the safety net that re-arms the loop if the chain ever breaks.
• Residual edge case: only if GitHub drops every scheduled tick for ~6 hours straight (longer than any gap observed so far) would there be a short pause at a hand-off, until the next tick fires and the loop resumes on its own.
• This runs the workflow almost continuously. On a public repo that's free (unlimited Actions minutes); on a private repo it would quickly exhaust the monthly free minutes, so keep the repo public or budget accordingly.
• The automated "goodmorning claude" goes to the /v1/messages API, not the Claude apps — so it will not appear in your Claude.ai / desktop conversation history. Its only visible effect is that your 5-hour usage window is started (visible on your usage/limits page while it's active).

Licence

MIT. See LICENSE.