Writing Hookify Rules

You define custom AI behavior rules for bash commands and file operations by creating a declarative markdown file with YAML frontmatter. This format specifies event-driven triggers, complex conditions, and regex patterns to govern agent actions.

Yes, you can block specific bash commands like rm -rf by creating a custom rule with a blocking action. The rule uses event-driven triggers and regex patterns to intercept the command and prevent execution before it happens.

Event-driven AI safety enforcement works by triggering predefined rules during specific agent events, such as user prompts, file edits, or bash execution. Rules evaluate complex conditions and regex patterns to either warn users or block operations proactively.

To prevent accidental file deletion, create a rule in a .local.md file within the .claude directory. Define the event type, match file operation patterns, and set a blocking action to stop the agent from deleting targeted directories.

Yes, the rule format supports regex patterns for matching complex conditions in user prompts. You can define declarative rules that evaluate user prompt text against regex patterns to trigger warning or blocking actions.

Rules are limited to the specific event triggers defined in the format: bash commands, file operations, agent stops, and user prompts. Complex conditional logic outside these events or beyond regex pattern matching capabilities cannot be enforced.