Anthropic API Integration

Relevant source files

The following files were used as context for generating this wiki page:

• src/validation/models/AnthropicApi.ts
• src/validation/models/AnthropicApi.test.ts
• src/config/Config.ts (Referenced via src/config/Config.test.ts)
• src/validation/validator.ts (Referenced via src/validation/validator.test.ts)
• src/validation/prompts/system-prompt.ts (Referenced via src/validation/models/AnthropicApi.ts)
• CLAUDE.md

The Anthropic API integration within TDD Guard provides a direct communication channel to Anthropic's language models to validate whether code changes adhere to Test-Driven Development (TDD) principles. It serves as one of the primary IModelClient implementations, enabling the system to analyze intercepted file operations, test results, and linting data to make "block" or "allow" decisions.

The integration is encapsulated in the AnthropicApi class, which utilizes the official @anthropic-ai/sdk. This module is responsible for constructing requests with specific system prompts, managing API credentials, and parsing the AI's structured JSON responses to determine TDD compliance.

Sources: CLAUDE.md:37-43, src/validation/models/AnthropicApi.ts:7-10

Architecture and Components

The integration follows an interface-driven design, ensuring that the validation logic remains decoupled from the specific AI provider.

The AnthropicApi Class

The AnthropicApi class implements the IModelClient interface. It is initialized with a Config object which provides the necessary API key and model version.

Communication Flow

When the validator component requires a decision, it invokes the ask method. The integration then formats a message for the Anthropic Messages API.

Implementation Details

Request Configuration

The integration enforces specific parameters to ensure consistent and high-quality TDD validation:

Parameter	Source/Value	Description
model	config.modelVersion	Defaults to claude-3-5-sonnet-20241022.
system	SYSTEM_PROMPT	Contains the core TDD rules and JSON formatting instructions.
max_tokens	1024	Limits the response length to ensure concise analysis.
messages	[{ role: 'user', content: prompt }]	Passes the dynamic context (code diffs, test output).

Sources: src/validation/models/AnthropicApi.ts:20-29, src/validation/models/AnthropicApi.test.ts:45-55, src/config/Config.test.ts:460-465

Response Processing

The API returns a MessageResponse object. The integration includes a private helper extractTextFromResponse to safely navigate the response structure. It specifically looks for the first content block and ensures it contains text. If the response is empty or lacks text content, the system throws an error which the validator catches to block the operation as a safety measure.

Sources: src/validation/models/AnthropicApi.ts:35-47, src/validation/validator.test.ts:153-158

Configuration Options

The behavior of the Anthropic integration is governed by environment variables and the Config class.

Environment Variable	Description	Default Value
TDD_GUARD_ANTHROPIC_API_KEY	The API key for Anthropic.	undefined
TDD_GUARD_MODEL_VERSION	Specific Claude model to use.	claude-3-5-sonnet-20241022
VALIDATION_CLIENT	Set to api to use this integration.	cli

Sources: src/config/Config.test.ts:200-210, src/config/Config.test.ts:285-300, src/config/Config.test.ts:455-465

Error Handling and Validation

The integration is designed to fail-safe. If the API call fails or returns unexpected data, the validator interprets this as a TDD violation or a system error that requires a "block" decision to prevent unvalidated code from being committed.

// Example of response extraction logic
function extractTextFromResponse(response: MessageResponse): string {
  if (response.content.length === 0) {
    throw new Error('No content in response')
  }

  const firstContent = response.content[0]
  if (!('text' in firstContent) || !firstContent.text) {
    throw new Error('Response content does not contain text')
  }

  return firstContent.text
}

Sources: src/validation/models/AnthropicApi.ts:35-47, src/validation/validator.test.ts:145-155

Conclusion

The Anthropic API Integration is a critical component of TDD Guard's validation engine. By leveraging high-capability models like Claude 3.5 Sonnet and strict system prompts, it provides the "intelligence" required to interpret complex TDD scenarios, such as distinguishing between minimal implementation stubs and over-implementation.