Claude Code · Codex

Stretch your Claude Code week, and keep working when the limit hits

Hooks, model routing, and a token-lean rules file designed to cut wasted usage, plus a step-by-step for falling back to Codex. $12 (plus VAT where applicable).

Get Usage Limit Saver: $12 (plus VAT where applicable)
Usage Limit Saver by Stack & Ship: Claude Code hooks and rules, with a Codex fallback. Real hook output: after 3 recorded failures of npm run build, the next identical run gets deny, "Running it again unchanged is unlikely to help." Documented example status line: 7d 41% | 5h 24% | ctx 8%.

What's inside

  • Hooks, designed to:
    • warn before a very large prompt, a resumed session whose prompt cache has likely expired with a big context, or a filling context window;
    • warn when your 5-hour or weekly plan usage passes a threshold you choose (Pro and Max plans, from Claude Code's own status-line data);
    • stop a loop of the same failing shell command (3 failures in 30 minutes by default);
    • stop a runaway turn or session with tool-call caps;
    • skip reads of lockfiles, dependency and build folders, and minified files, and ask for a line range on very large files. Each hook is designed to fail open: if it errors, Claude Code carries on. Every case and setting is documented.
  • Model routing with documented settings only: an example settings file (main model, effort, subagent default), plus an Explore subagent and a quick-edit subagent on a smaller model.
  • Token-lean rules: a CLAUDE.md snippet (with compact instructions) and an AGENTS.md.
  • Codex fallback step-by-step: AGENTS.md, a handoff template, and a /handoff skill so a fresh Codex or Claude Code session can pick up where you stopped.
  • Weekly budget checklist and a usage-log script (Node, no dependencies) that summarises your 5-hour and weekly percentages over the past days.
  • Installer designed to merge into your existing settings without overwriting them, with a backup and an uninstall option.

28 files: hooks and scripts, installer, 2 subagents, 1 skill, 2 rules files, 6 guides (hooks, model routing, Codex fallback, weekly checklist, handoff template, evidence), tests, README, changelog, and licence.

Install

  1. Unzip. To preview first, run node install.mjs --project /path/to/your/repo --dry-run.
  2. Run node install.mjs --project /path/to/your/repo --extras (or node install.mjs --user --extras for all projects).
  3. Paste rules/CLAUDE.token-lean.md into your CLAUDE.md. If you use Codex, copy rules/AGENTS.md to your repo root.
  4. Restart Claude Code and run /hooks to check.

To remove it, run the installer with --uninstall. Needs Node 20 or later. No npm install.

Tested

  • 48 automated tests pass (hook cases, status line and report, installer merge and uninstall, fail-safe) on Node 20 and Node 24.
  • 16 live checks with the real Claude Code CLI 2.1.302 against a local mock API: loop stop, turn caps, large-prompt warning and confirm mode, read guard, missing-script fail-safe, model routing, subagents and skill loading, and side by side with the Stack & Ship Guardrails Pack 1.2.0.
  • 5 live checks with codex-cli 0.164.0: AGENTS.md reaches the model's prompt, and the config keys and resume commands in the guide exist.
  • Tested on Linux only so far, not on macOS or Windows. Not tested against a live Anthropic or OpenAI account.

What this doesn't do

  • It doesn't raise your plan's limits or change how they're counted. It's designed to reduce waste in your own sessions.
  • It makes no promise about how much extra usage you'll get. That depends on your work, your plan, and how the vendors count usage, which can change.
  • The usage-log script records only what Claude Code passes to your status line and hooks on your machine (plan usage percentages, reset times, context use, tool calls). It doesn't read transcripts or call any API, and it isn't an official meter: /usage in Claude Code is the source of truth. Plan percentages appear only on Pro and Max plans, in interactive sessions.
  • If you already have your own status line, the installer leaves it as is, and the usage log and plan-usage warnings then need a small manual step (see docs/HOOKS.md).
  • The Codex step-by-step needs your own Codex access. Codex isn't included.
  • Hooks are designed to warn or stop; you stay in control and can change or turn off any of them.

Who it's for

Solo developers who keep running out of weekly Claude Code usage before the week ends.

FAQ

Will this give me more usage each week?
It's designed to cut common sources of waste: oversized context, retry loops, and big models doing small jobs. We don't claim a specific amount, because it depends on how you work.
Do I need Codex?
Only for the fallback step-by-step. The hooks, rules, checklist, and script work with Claude Code on their own.
Will it change my existing setup?
The installer is designed to merge into your existing settings.json, not overwrite it. It saves a backup first, keeps your hooks, permissions, and other settings, adds its own hook entries once (running it again doesn't duplicate them), never sets your model or permissions, keeps your status line if you have one, and stops without changes if the file isn't valid JSON. --uninstall removes only its own entries.
Does it work with the Guardrails Pack?
Yes. We tested both installed in the same project with Guardrails Pack 1.2.0: Guardrails' hooks and deny rules stayed in place, Guardrails still denied a force push, and Usage Limit Saver's loop stop still worked.

Disclosure: this is our own paid product.
Usage Limit Saver is $12 (plus VAT where applicable), one payment.

Get Usage Limit Saver: $12 (plus VAT where applicable)

Anthropic, Claude and Claude Code are trademarks of Anthropic. OpenAI and Codex are trademarks of OpenAI. Node.js is a trademark of the OpenJS Foundation. npm is a trademark of npm, Inc. Other product names are trademarks of their respective owners. Names are used only to identify products and compatibility. Stack & Ship is not affiliated with or endorsed by any of them.