Writing Hookify Rules

To create custom AI guardrails, you define Hookify rules using YAML frontmatter and message body content in a markdown file. This system triggers specific agent behaviors or warnings based on events like bash commands, file edits, agent stops, or user prompts.

You can target four event types when defining AI agent behavior rules: bash, file, stop, and prompt. These events allow you to intercept and evaluate agent actions, triggering warnings or blocks for specific conditions like dangerous command execution.

You block dangerous shell commands by creating a Hookify rule that targets the bash event type. Using Python regex for pattern matching in your rule conditions, you can detect commands like `rm -rf` and output a warning message in the rule body to prevent execution.

Yes, you can use Python regex for pattern matching to build advanced conditions in AI guardrails. This allows you to evaluate multiple conditions simultaneously, checking text content and file paths to warn about sensitive edits like committing API keys into a `.env` file.

A Hookify rule file is a markdown file, such as `.claude/hookify.warn-dangerous-rm.local.md`, containing YAML frontmatter to define event types and conditions, followed by a message body that provides the warning or constraint output for the AI agent.

No external dependencies are required to start writing AI guardrails with Hookify. You only need to create a local markdown rule file with the correct YAML frontmatter syntax and message body content to enforce your custom agent constraints.