mockgateway

type ErrorInject struct {
	Status int `yaml:"status"`
	Times  int `yaml:"times"`
}

type Recorded struct {
	// Endpoint is the request path (/v1/chat/completions or /v1/messages).
	Endpoint string
	// Provider is the ?provider= query parameter sent by the SDK.
	Provider string
	// Model is the request-body model (provider prefix already stripped).
	Model string
	// Scenario is the matched scenario name; empty when only the fallback applied.
	Scenario string
	// Step is the assistant-message count at the time of the request.
	Step int
	// Stream reports whether the request asked for SSE.
	Stream bool
	// Body is the full decoded request for /v1/chat/completions.
	Body sdk.CreateChatCompletionRequest
	// MessagesBody is the full decoded request for /v1/messages.
	MessagesBody *sdk.CreateMessagesRequest
}

type Scenario struct {
	// Name uniquely identifies the scenario in recordings and logs.
	Name string `yaml:"name"`
	// Match is a Go regular expression tested (unanchored) against the latest
	// user message of each request that is not an injected <system-reminder>.
	Match string `yaml:"match"`
	// Turns are the scripted assistant responses, indexed by the number of
	// assistant messages following the matched user message.
	Turns []Turn `yaml:"turns"`
	// contains filtered or unexported fields
}

type ScenarioFile struct {
	// Fallback is rendered when no scenario matches the prompt or when a
	// matched scenario has no turn left for the current step.
	Fallback Turn `yaml:"fallback"`
	// Scenarios are evaluated in file order; the first regex match wins.
	Scenarios []Scenario `yaml:"scenarios"`
}

type Server struct {
	Model string
	// contains filtered or unexported fields
}

type StallInject struct {
	Times   int  `yaml:"times"`
	Connect bool `yaml:"connect"`
}

type ToolCall struct {
	// Name is the tool name as registered in the CLI (e.g. Read, Grep, Bash).
	Name string `yaml:"name"`
	// Args is marshaled into the tool call's JSON arguments string.
	Args map[string]any `yaml:"args"`
	// contains filtered or unexported fields
}

type Turn struct {
	// Content is the assistant text, streamed in ChunkSize-rune fragments.
	Content string `yaml:"content"`
	// Reasoning is streamed as reasoning_content deltas before Content.
	Reasoning string `yaml:"reasoning"`
	// ToolCalls all land in this single assistant turn.
	ToolCalls []ToolCall `yaml:"tool_calls"`
	// Usage defaults to 10 prompt / 5 completion tokens when nil.
	Usage *Usage `yaml:"usage"`
	// ChunkSize is the fragment size in runes for streamed text (default 16).
	ChunkSize int `yaml:"chunk_size"`
	// DelayMs sleeps before each SSE frame (streaming) or once before the
	// body (sync), aborting early when the client disconnects.
	DelayMs int `yaml:"delay_ms"`
	// Error, when set, replaces the turn with an HTTP error for the first
	// Times matching requests (-1 means every request).
	Error *ErrorInject `yaml:"error"`
	// Stall, when set, makes the first Times streaming requests hang after
	// the initial role delta: the connection stays open but no further
	// frames arrive until the client disconnects. Exercises the CLI's
	// stalled-stream reconnect path.
	Stall *StallInject `yaml:"stall"`
	// Malformed emits one non-JSON data: frame early in the stream.
	Malformed bool `yaml:"malformed"`
}

type Usage struct {
	PromptTokens     int64 `yaml:"prompt_tokens"`
	CompletionTokens int64 `yaml:"completion_tokens"`
	CachedTokens     int64 `yaml:"cached_tokens"`
	CacheWriteTokens int64 `yaml:"cache_write_tokens"`
}