Anthropic Provider

The shieldAnthropic wrapper adds transparent security to your existing Anthropic client. It intercepts every messages.create call to harden the system prompt, detect injections in user messages, and sanitize leaked content from responses.

Usage

import Anthropic from "@anthropic-ai/sdk";
import { shieldAnthropic } from "@zeroleaks/shield/anthropic";

const client = shieldAnthropic(new Anthropic(), {
  systemPrompt: "You are a support agent...",
});

const response = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  system: "You are a support agent...",
  messages: [{ role: "user", content: userInput }],
  max_tokens: 1024,
});

How It Works

On every call to messages.create, Shield:

1. Clones the params object (never mutates your original)
2. Hardens the system field if it is a string (unless harden: false)
3. Scans every user message for injection patterns, supporting both string content and content block arrays (unless detect: false)
4. Calls the original Anthropic API
5. Sanitizes the first text block in the response for leaked system prompt fragments (unless sanitize: false)

Options

Content Block Support

Anthropic messages can contain content blocks (arrays of {type, text} objects) instead of plain strings. Shield extracts text from these blocks for detection:

const response = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  system: "You are a support agent...",
  messages: [
    {
      role: "user",
      content: [
        { type: "text", text: "Help me with this document:" },
        { type: "text", text: documentContent },
      ],
    },
  ],
  max_tokens: 1024,
});

Both text blocks are scanned for injection patterns.