Inject Context into Every Prompt with the UserPromptSubmit Hook

The UserPromptSubmit hook fires every time you submit a prompt, before Claude processes it. You can use it to automatically inject additional context, enforce guardrails, or block certain prompts entirely.

Here's a simple hook that appends the current git branch and recent commits to every prompt:

#!/bin/bash
# .claude/hooks/inject-context.sh

BRANCH=$(git branch --show-current 2>/dev/null)
RECENT=$(git log --oneline -3 2>/dev/null)

cat << EOF
{
  "hookSpecificOutput": {
    "hookEventName": "UserPromptSubmit",
    "additionalContext": "Current branch: $BRANCH\nRecent commits:\n$RECENT"
  }
}
EOF

Register it in .claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": ".claude/hooks/inject-context.sh"
          }
        ]
      }
    ]
  }
}

You can also block prompts by returning a decision: "block" response with a reason. This is useful for enforcing team policies — for example, blocking prompts that ask Claude to skip tests or bypass code review steps.

The hook receives the user's prompt text as prompt in the JSON input on stdin, so you can inspect and respond to specific content.

UserPromptSubmit hooks are a clean way to add invisible context to every single Claude interaction.

via Claude Code Docs — Hooks