@aegis-sdk/anthropic

@aegis-sdk/anthropic

Anthropic Claude SDK adapter for Aegis prompt injection defense. Scans messages, monitors streams, and validates tool_use blocks in Anthropic's content block format.

Part of the Aegis.js prompt injection defense toolkit.

Installation

npm install @aegis-sdk/anthropic @aegis-sdk/core @anthropic-ai/sdk

Quick Start

Wrap the Anthropic client for automatic protection on every call:

import Anthropic from '@anthropic-ai/sdk';
import { Aegis } from '@aegis-sdk/core';
import { wrapAnthropicClient } from '@aegis-sdk/anthropic';

const aegis = new Aegis({ policy: 'strict' });
const client = wrapAnthropicClient(new Anthropic(), aegis);

// Messages are scanned before sending.
// Streaming responses are monitored in real-time.
// tool_use blocks are validated against the policy.
const response = await client.messages.create({
  model: 'claude-sonnet-4-20250514',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello!' }],
});

Or scan messages manually:

import { Aegis } from '@aegis-sdk/core';
import { guardMessages } from '@aegis-sdk/anthropic';

const aegis = new Aegis({ policy: 'strict' });

const messages = [
  {
    role: 'user' as const,
    content: [
      { type: 'text' as const, text: userInput },
    ],
  },
];

// Throws AegisInputBlocked if injection is detected
const safe = await guardMessages(aegis, messages);

API

wrapAnthropicClient(client, aegis, options?)

Proxy the Anthropic client to automatically guard all messages.create() calls. Input messages are scanned before sending, streaming responses are monitored, and tool_use content blocks are validated against the Aegis policy. All other client methods pass through unchanged.

guardMessages(aegis, messages, options?)

Scan an array of AnthropicMessageParam[] for prompt injection. Handles Anthropic's message format -- string content, arrays of content blocks (text, tool_use, tool_result, image), and nested text within tool_result blocks. Optionally validates tool_use blocks against the policy (enabled by default). Returns the original messages if safe, throws AegisInputBlocked if blocked.

createStreamTransform(aegis)

Create a TransformStream<string, string> for monitoring extracted text deltas. Feed text_delta values from Anthropic streaming events through this transform for real-time output scanning.

getAuditLog(aegis)

Convenience accessor for the Aegis audit log.

Options

guardMessages and the client wrapper accept AnthropicGuardOptions:

• scanStrategy -- 'last-user' (default), 'all-user', or 'full-history'
• validateToolUse -- Whether to validate tool_use blocks against the policy (default: true)

Learn More

• Documentation
• GitHub

License

MIT