Class AnthropicModerationCapabilityImpl

Adapts Anthropic moderation responses into ProviderPlaneAI's normalized moderation artifact surface.

Anthropic does not expose a native moderation endpoint, so this adapter prompts Claude for deterministic JSON and normalizes the parsed result.

Implements

ModerationCapability<ClientModerationRequest, NormalizedModeration[]>

Index

Constructors

Methods

Constructors

constructor

new AnthropicModerationCapabilityImpl(
provider: BaseProvider,
client: Anthropic,
): AnthropicModerationCapabilityImpl
Creates a new Anthropic moderation capability adapter.
Parameters
- provider: BaseProvider
  Parent provider instance used for initialization checks and merged config access.
- client: Anthropic
  Initialized Anthropic SDK client.
Returns AnthropicModerationCapabilityImpl
- Defined in src/providers/anthropic/capabilities/AnthropicModerationCapabilityImpl.ts:85

Methods

moderation

moderation(
    request: AIRequest<ClientModerationRequest>,
    _executionContext?: MultiModalExecutionContext,
    signal?: AbortSignal,
): Promise<AIResponse<NormalizedModeration[]>>
Executes an Anthropic moderation request.

Responsibilities:
- validate moderation input
- resolve merged model/runtime options
- execute one Claude moderation prompt per input string
- parse deterministic JSON output into normalized moderation artifacts
- aggregate token usage and attach provider/request metadata
Parameters
- request: AIRequest<ClientModerationRequest>
  Unified moderation request envelope.
- Optional_executionContext: MultiModalExecutionContext
  Optional execution context. Unused directly in this adapter.
- Optionalsignal: AbortSignal
  Optional cancellation signal.
Returns Promise<AIResponse<NormalizedModeration[]>>
Provider-normalized moderation artifacts.

Throws
When input is invalid, aborted, or Claude returns no text moderation payload.
Implementation of ModerationCapability.moderation
- Defined in src/providers/anthropic/capabilities/AnthropicModerationCapabilityImpl.ts:106