Skip to main content

Setting a spending limit on a Claude Code session

An agent that runs on its own stops when the task is done, not when your budget is. That’s fine for a quick question, but it’s a different story for a refactoring job in CI or a scheduled run overnight.

So, the question becomes: how do you put a ceiling on a Claude Code session?


Remark: Strictly speaking you can’t set a token limit. Claude Code caps on spend (tokens × price), so that’s what we’ll use.

The naive approach

Just run it and see what happens:

claude -p "Run the test suite, fix the failing tests in src/ and commit the fix"

There is no limit by default. Claude keeps going until it thinks it’s finished, and you find out the cost afterwards.

Two flags can help

Claude Code has two flags that act as a safety net in print mode (-p):

claude -p "Run the test suite, fix the failing tests in src/ and commit the fix" \
  --max-budget-usd 3.00 \
  --max-turns 30
  • --max-budget-usd stops the run once the estimated spend on API calls reaches the amount.
  • --max-turns limits the number of agentic turns. When the limit is reached, the run exits with an error, so your script can branch on the exit code.

I typically use both. A runaway loop that keeps retrying a failing fix hits the turn limit long before it gets expensive, and the budget flag covers the case where every turn is just very large.

Remark: The budget is a soft cap. It’s checked against a client-side estimate that can differ from your actual bill, and spend can pass the cap. Best to leave some headroom.

What about interactive sessions?

As far as I know, there is no equivalent hard cap for an interactive session. What you can do:

  • Check the running cost with /cost.
  • Look at what is filling your context window with /context.
  • Use /compact to summarize a long session, or /clear when you switch tasks.
  • When you’re billed through the API, set a spend limit on the workspace in the Anthropic Console as the backstop.

That’s it! Set both flags on every unattended run and treat the result as a safety net, not an exact meter.

More information

Popular posts from this blog

Podman– Command execution failed with exit code 125

After updating WSL on one of the developer machines, Podman failed to work. When we took a look through Podman Desktop, we noticed that Podman had stopped running and returned the following error message: Error: Command execution failed with exit code 125 Here are the steps we tried to fix the issue: We started by running podman info to get some extra details on what could be wrong: >podman info OS: windows/amd64 provider: wsl version: 5.3.1 Cannot connect to Podman. Please verify your connection to the Linux system using `podman system connection list`, or try `podman machine init` and `podman machine start` to manage a new Linux VM Error: unable to connect to Podman socket: failed to connect: dial tcp 127.0.0.1:2655: connectex: No connection could be made because the target machine actively refused it. That makes sense as the podman VM was not running. Let’s check the VM: >podman machine list NAME         ...

Cache stampede: when our cache turned against us

While investigating some performance issues, we ran into an ASP.NET Core API that cached a fairly expensive aggregation query for 60 seconds. Under normal load, that was fine: one request rebuilds the cache, everyone else reads from it. Under peak load, dozens of requests would arrive in that same expiry window, all see a cache miss, and all fire the same expensive query in parallel. The database didn't like that. That was the moment when our caching layer stopped helping and started hurting. A burst of requests comes in at the same time, all miss the cache, and all go hammer the database or the downstream API at once. That's a cache stampede . The cache was supposed to protect our backend, and for a few hundred milliseconds it did the opposite. Why this happens IMemoryCache.GetOrCreate (and its async sibling) looks like it protects you, but it doesn't add any locking on its own. Look at the naive version: public async Task<Report> GetReportAsync(string key) ...

VS Code Planning mode

After the introduction of Plan mode in Visual Studio , it now also found its way into VS Code. Planning mode, or as I like to call it 'Hannibal mode', extends GitHub Copilot's Agent Mode capabilities to handle larger, multi-step coding tasks with a structured approach. Instead of jumping straight into code generation, Planning mode creates a detailed execution plan. If you want more details, have a look at my previous post . Putting plan mode into action VS Code takes a different approach compared to Visual Studio when using plan mode. Instead of a configuration setting that you can activate but have limited control over, planning is available as a separate chat mode/agent: I like this approach better than how Visual Studio does it as you have explicit control when plan mode is activated. Instead of immediately diving into execution, the plan agent creates a plan and asks some follow up questions: You can further edit the plan by clicking on ‘Open in Editor’: ...