Railgun Security Manager

Railgun Security Manager

Railgun protects your Claude Code sessions by intercepting tool calls and blocking:

• Secrets: AWS keys, GitHub tokens, OpenAI keys, private keys
• Dangerous commands: rm -rf /, fork bombs, disk writes
• Protected paths: .env, .ssh/, *.pem, credentials files
• Data exfiltration: pastebin, ngrok, webhook capture sites

Commands

Lint Configuration

Validate your railgun.toml configuration file:

# Find and run the appropriate binary
BINARY="${CLAUDE_PLUGIN_ROOT}/bin/rg-$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m | sed 's/x86_64/x64/' | sed 's/aarch64/arm64/')"
"$BINARY" lint --config ./railgun.toml

Test a Tool Input

Test how a specific tool input would be evaluated by the policy:

# Test a Bash command
echo '{"tool_name":"Bash","tool_input":{"command":"ls -la"}}' | "$BINARY" hook --config ./railgun.toml

# Test a file read
echo '{"tool_name":"Read","tool_input":{"file_path":"./src/main.rs"}}' | "$BINARY" hook --config ./railgun.toml

Exit code 0 = allowed, exit code 2 = blocked.

Create Configuration

Copy the example configuration to customize:

cp "${CLAUDE_PLUGIN_ROOT}/railgun.example.toml" ./railgun.toml

Configuration

Railgun looks for configuration in this order:

1. ./railgun.toml (project-level)
2. ~/.config/railgun/railgun.toml (user-level)
3. Plugin default (read-only)

Policy Modes

• strict (default): Block dangerous actions
• monitor: Log only, don't block (for testing)

Getting Help

If Railgun blocks something you need:

1. Check which pattern matched in the block reason
2. Add an allow pattern to override the block
3. Or disable the specific scanner in config