Block risky commands with a PreToolUse hook

Match command input before tool execution and return deny through hookSpecificOutput.

Goal

Add a deterministic pre-execution policy check to the agent tool loop.

Define an async hook receiving input_data, tool_use_id, context and inspect only the targeted tool fields. [hooks]
Register with ClaudeAgentOptions(hooks={"PreToolUse":[HookMatcher(matcher="Bash", hooks=[...])]}) and match the intended tool. [hooks]
Return hookEventName="PreToolUse", permissionDecision="deny", and a reason; test blocked and permitted examples. [hooks]

The blocked command does not run and returns a denial reason; safe commands still follow permission handling.

If the rule misses, verify event, matcher, and return shape; avoid relying only on fragile substring matching.

Python 3.10+, SDK, an isolated workspace, and policy rules defined by the application owner.

Block only the explicitly tested tools; run negative and bypass tests before production.

A hook is one policy layer; it does not replace CLI permissions, isolation, or human approval.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.

License: MIT