easy-ui-mcp: Skill for Claude Code

Role: Git Safety Hook Installer

Installs a PreToolUse hook that intercepts and blocks destructive git commands before Claude runs them — a guardrail for the isolated worktrees sub-agents work in (Stage 3). Complements update-config and fewer-permission-prompts in the Stage 1 one-time setup checklist.

What gets blocked

git push (all variants incl. --force), git reset --hard, git clean -f / -fd, git branch -D, git checkout . / git restore .. When blocked, Claude is told it does not have authority to run the command.

Steps

1. Ask scope

Install for this project only (.claude/settings.json) or all projects (~/.claude/settings.json)?

2. Place the hook script

The bundled script lives at .claude/skills/git-guardrails-claude-code/scripts/block-dangerous-git.sh. Copy it to:

• Project: .claude/hooks/block-dangerous-git.sh
• Global: ~/.claude/hooks/block-dangerous-git.sh

Then chmod +x it.

3. Register the hook (via the update-config skill)

Add to the chosen settings file, merging into any existing hooks.PreToolUse array — never overwrite other settings:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          { "type": "command", "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/block-dangerous-git.sh" }
        ]
      }
    ]
  }
}

(Global scope: use ~/.claude/hooks/block-dangerous-git.sh as the command.)

4. Ask about customization

Ask whether to add/remove patterns from the blocked list; edit the copied script accordingly.

5. Verify

echo '{"tool_input":{"command":"git push origin main"}}' | .claude/hooks/block-dangerous-git.sh

Should exit code 2 and print a BLOCKED message to stderr.

Note for this framework

The Supervisor commits/pushes only when the user explicitly asks. This hook enforces that boundary mechanically — if the user later authorizes a push, they run it themselves (or temporarily disable the hook), keeping destructive git an explicit human decision.