/install

Install Prompt Injection Defender

Install the prompt injection defender hooks into your Claude Code project.

Instructions

Guide me through installing the prompt injection defender. Ask me:

1. **Which implementation?** Python (uv) or TypeScript (bun) 2. **Installation scope?** Project-only or user-wide 3. **Which tools to monitor?** All recommended tools or let me select

Installation Steps

After gathering my preferences:

1. **Check prerequisites**

• For Python: Verify `uv` is installed (`which uv`)
• For TypeScript: Verify `bun` is installed (`which bun`)

2. **Copy hook files**

• Copy the appropriate hook implementation to my project or user config
• For project: `.claude/hooks/prompt-injection-defender/`
• For user-wide: `~/.claude/hooks/prompt-injection-defender/`

3. **Copy patterns.yaml**

• Place `patterns.yaml` alongside the hook script
• This file contains all detection patterns

4. **Update settings configuration**

• **IMPORTANT**: Check if `.claude/settings.local.json` exists first
• If `settings.local.json` exists, merge hooks into THAT file (it takes precedence over `settings.json`)
• If only `settings.json` exists, merge hooks there
• Use the template from `python-settings.json` or `typescript-settings.json`
• Adjust paths based on installation scope
• For user-wide: update `~/.claude/settings.json`

**Settings precedence (highest to lowest):** 1. `.claude/settings.local.json` (local project - overrides all below) 2. `.claude/settings.json` (project settings) 3. `~/.claude/settings.json` (user settings)

5. **Restart Claude Code**

• **CRITICAL**: Hooks are captured at session startup
• Tell the user they MUST restart Claude Code for hooks to take effect
• Exit the current session and start a new one

6. **Verify installation**

• After restart, run a quick test to ensure the hook is working
• Read a test file with known injection patterns
• Use `/hooks` command to verify hooks are loaded

Recommended Tool Configuration

Monitor these tools for prompt injection:

• `Read` - File contents
• `WebFetch` - Web page contents
• `Bash` - Command outputs
• `Grep` - Search results
• `Task` - Agent outputs
• `mcp__*` - All MCP tool outputs

After Installation

**IMPORTANT**: Tell the user to restart Claude Code! Hooks only load at startup.

Explain what will happen after restart:

• When Claude reads files, fetches web pages, or runs commands
• The hook will scan the output for injection patterns
• If patterns are detected, Claude will see a warning
• Processing continues but Claude is warned to be cautious

The user can verify hooks are active by running `/hooks` in Claude Code.

Reference Files

• Skill location: `.claude/skills/prompt-injection-defender/`
• Python hook: `hooks/defender-python/post-tool-defender.py`
• TypeScript hook: `hooks/defender-typescript/post-tool-defender.ts`
• Patterns: `patterns.yaml`
• Settings templates: `python-settings.json` / `typescript-settings.json`