vigiles API reference
    Preparing search index...

    Interface ModelMock

    ModelMock — the model-mock PORT (transport axis). A harness test tier runs the real agent against a fake, scripted model server so it needs no API key and is deterministic. This descriptor captures what the mock's WIRE FORMAT is: the API dialect it speaks, the endpoint that consumes a scripted turn, and the optional token-count endpoint a client probes. The research (research/harness-landscape.md) found the two concrete formats differ sharply — Claude Code speaks Anthropic Messages SSE at /v1/messages; Codex speaks OpenAI Responses SSE at /v1/responses — which is exactly the second implementation that makes this port designable.

    The HTTP server + per-event flush mechanics live in the harness's mock module (startMock for Claude Code); a second harness implements the same contract with its own renderer. The Claude Code descriptor is claudeCodeModelMock in src/adapters/claude-code/model-mock.ts.

    interface ModelMock {
        countTokensEndpoint?: string;
        modelEndpoint: string;
        name: string;
        wireApi: string;
    }
    Index

    Properties

    countTokensEndpoint? modelEndpoint name wireApi

    Properties

    countTokensEndpoint?: string

    Token-count endpoint substring a client probes (e.g. "count_tokens"); omit if none.

    modelEndpoint: string

    URL substring identifying a turn-consuming model request (e.g. "/v1/messages").

    name: string

    Stable identifier, e.g. "claude-code".

    wireApi: string

    The model API wire format, e.g. "anthropic-messages" or "openai-responses".

    Settings

    Member Visibility

    On This Page

    Properties