Skip to main content

mekara hook

Hook handlers for Claude Code integration. These commands are invoked by Claude Code hooks configured in ~/.claude/settings.json.

Installation

These hooks are automatically configured when you run mekara install hooks or mekara install. You typically don't need to invoke these commands manually.

Usage​

mekara hook reroute-user-commands
mekara hook reroute-agent-commands
mekara hook auto-approve

Subcommands​

mekara hook reroute-user-commands​

Intercepts user input to detect /command syntax and routes script execution through the MCP server.

Input: Reads JSON from stdin with the format:

{
  "prompt": "/test/random arg1 arg2"
}

Output:

  • For compiled scripts: Outputs instructions telling Claude to call mcp__mekara__start
  • For natural language scripts: Outputs the full script content with $ARGUMENTS substituted

Configuration: Triggered by the UserPromptSubmit hook in ~/.claude/settings.json.

See reroute-user-commands implementation for detailed input/output schemas and script resolution logic.

mekara hook reroute-agent-commands​

Intercepts the Skill tool to ensure compiled mekara scripts execute via MCP instead of the Skill tool.

Input: Reads JSON from stdin with the format:

{
  "tool_name": "Skill",
  "tool_input": {
    "skill": "test:random",
    "args": "arg1 arg2"
  }
}

Output:

  • For compiled scripts: Returns a deny permission decision with instructions to use MCP
  • For non-mekara skills: Returns exit code 0 with no output (allows Skill tool to proceed)

Configuration: Triggered by the PreToolUse hook with matcher "Skill" in ~/.claude/settings.json.

See reroute-agent-commands implementation for detailed permission decision logic.

mekara hook auto-approve​

Automatically approves most Claude Code tool invocations to reduce permission prompts.

Input: Reads JSON from stdin with the format:

{
  "tool_name": "Bash",
  "tool_input": {
    "command": "ls -la"
  }
}

Output:

  • For safe operations: Returns allow permission decision
  • For dangerous operations (rm, git commit): Returns exit code 0 with no output (requires user confirmation)

Configuration: Triggered by the PreToolUse hook with matcher "" (matches all tools) in ~/.claude/settings.json.

Workaround for Permissions Bugs

This hook exists as a workaround for Claude Code permissions bugs. It provides a minimal safety net while reducing friction.

For more comprehensive permissions control with fine-grained rules, see claude-code-permissions-hook.

See auto-approve implementation for detailed permission decision logic.

Hook Configuration​

The mekara install hooks command creates this configuration in ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "mekara hook reroute-user-commands"
          }
        ]
      }
    ],
    "PreToolUse": [
      {
        "matcher": "Skill",
        "hooks": [
          {
            "type": "command",
            "command": "mekara hook reroute-agent-commands"
          }
        ]
      },
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "mekara hook auto-approve"
          }
        ]
      }
    ]
  }
}

The matcher field filters which tools trigger the hook:

  • Empty string "" matches all tools/prompts
  • "Skill" matches only the Skill tool

See Also​