Agent Spend / Token Survival Kit

A set of worksheets that help you cap, notice and review what your coding agents spend, so a runaway loop or a leaked key is a small problem instead of a surprise bill.

What is inside

• Spend checklist (pass / fail worksheet): spend caps, kill-loop habits, context hygiene, and key hygiene and provider alerts. You fill in your own caps, turn limits and alert levels.
• Model-routing sheet: a short decision tree and table for when a cheaper model is enough and when to pay for a smarter one, filled in for your own stack.
• Weekly bill review: a one-page template for spend by tool, what burned the most, a loop and hygiene audit, and at most three rules for next week.
• README: a first-week plan of about 30 to 60 minutes to set it all up.

Everything is plain Markdown you edit yourself, in whatever editor or notes tool you already use.

Spend rules and the weekly review

The checklist turns into a short set of rules you write down once, such as your monthly hard cap, a maximum number of agent turns per task, and the level at which each provider should alert you. The weekly review is the repeatable part: a scheduled look at your latest bills to check whether the rules held, ending with at most three rules for the next week.

Why this goes beyond free advice

Much of the underlying advice is free, and we link to it. The vendors document their own switches, for example Claude usage-credit limits and the turn and budget flags in the Claude Code CLI reference, and the free spend stories on this site cover several common failures one at a time. If you only need one of those fixes, use the free page.

What the Kit adds is structure across tools and over time:

• One place to write down your caps for every tool and key, instead of finding them in six dashboards after an alert fires.
• A pass / fail checklist that shows which limits you have not set, including the ones that only alert and do not stop anything.
• A weekly review so the numbers get looked at on a schedule, not only after a bad month.
• Tool-agnostic wording, so it still applies when you change editors, models or plans.

Who it is for

• Solo builders and small teams who run Cursor, Claude Code, Codex or similar agents on a plan or with their own API keys.
• People who have been surprised by a usage bill, a plan limit, or a loop that kept running.
• Anyone who prefers a checklist they run themselves to another dashboard.

Probably not for you if you want software that watches your accounts for you, if your company already has enforced spend controls, or if the free stories above already cover your one problem.

What it is not, and what we do not promise

• It is not a dashboard, a plugin or a service. It does not connect to your accounts.
• We do not guarantee any savings. We have not measured the Kit's effect on anyone's bill, and prices, plans and tools change.
• It does not give you access to any tool, discount or shared account.

Terms

License: personal or single-business use. No resale, redistribution or white-labeling without a separate license.

Refunds: REFUND_POLICY_TEXT

Staging note (not shown in a release build): the refund wording above is the placeholder REFUND_POLICY_TEXT. The earlier internal draft (no refunds after download, except a corrupt file within 7 days) has not been confirmed by the user.

Free reading that goes with it

• From December, ChatGPT and Codex share one usage allowance on Plus, Pro, and Business
• Grandfathered ChatGPT Pro 200 keeps the old included allowance only through October 29
• OpenCode Console breach: attackers ran up charges on users' stored AI provider keys
• Hitting a Claude plan limit can flip to pay-as-you-go usage credits, if you enabled them
• Headless claude -p has no turn limit by default; --max-budget-usd is the documented spend cap
• An alert is an email, a cap is a breaker: one builder's cap that fired and the five-layer stack he recommends