Review the Tool Calls That Write, Spend, or Leave the Machine

Wed, 07 Oct 2026 14:04:27 +0000

A person should see the calls that change something. Reads can run when a policy already allows that tool and those arguments.

The failure mode

Reviewing every tool call does not scale, and reviewing none of them leaves a coding agent free to write files, spend money, send messages, or leave the machine. A review after the call is a transcript of damage that already happened.

The check

ThumbGate checks the tool name and the arguments before the call runs. The gate shows one line of intent for a write, a payment, a message, or a command that leaves the machine: the command, the path, or the destination. A read or a local lookup proceeds when the policy already allows that tool and those arguments.

function reviewBeforeRun(toolName, args) {
  const command = String(args.command || '');
  const leaves = toolName === 'Bash' && /\b(curl|ssh|scp|git\s+push)\b/.test(command);
  const writes = toolName === 'Write' || toolName === 'Edit';
  const sends = toolName === 'SendMessage' || toolName === 'WebFetch';
  if (leaves || writes || sends) {
    return { decision: 'review', show: command || args.path || args.url || toolName };
  }
  return { decision: 'allow' };
}

The check belongs in the PreToolUse hook, on the tool name and the arguments, before a shell or network call runs.

Try it

Install with npx thumbgate init.

ThumbGate Pro is $19 a month or $149 a year per operator.

Pricing: https://thumbgate.ai/pricing

Checkout: https://thumbgate.ai/go/pro?utm_source=coderlegion&utm_medium=community&utm_campaign=pretool-hook

Repo: https://github.com/IgorGanapolsky/ThumbGate

Check the tool call before a coding agent runs it

Mon, 05 Oct 2026 16:36:49 +0000

A system prompt asks the model to behave. The tool call still goes out when the model ignores that ask.

ThumbGate sits in front of the call. The PreToolUse hook reads the command before it runs and can block it. A thumbs-down on a bad call becomes a rule, so the same call does not get through the next time.

Install on a repo with npx thumbgate init.

Pro is $19 a month or $149 a year per operator.

https://thumbgate.ai/pricing

https://thumbgate.ai/go/pro?utm_source=coderlegion&utm_medium=community&utm_campaign=pretool-hook

The longer piece on why the prompt is the wrong layer: https://coderlegion.com/28973/why-system-prompts-wont-save-your-ai-agent-building-architectural-injection-immunity

Coder Legion - Recent posts tagged pretooluse

Review the Tool Calls That Write, Spend, or Leave the Machine

The failure mode

The check

Try it

Check the tool call before a coding agent runs it