Documentation
¶
Index ¶
- func BuildAgentMessagesFromEntries(entries []domain.ConversationEntry) []sdk.Message
- type Options
- type Runner
- func (r *Runner) HandleChatChunk(msg domain.ChatChunkEvent) tea.Cmd
- func (r *Runner) HandleChatComplete(msg domain.ChatCompleteEvent) tea.Cmd
- func (r *Runner) HandleChatError(msg domain.ChatErrorEvent) tea.Cmd
- func (r *Runner) HandleChatStart(_ domain.ChatStartEvent) tea.Cmd
- func (r *Runner) HandleOptimizationStatus(event domain.OptimizationStatusEvent) tea.Cmd
- func (r *Runner) SetPendingRestoration(originalModel string)
- func (r *Runner) Start(holder domain.BashDetachChannelHolder) tea.Cmd
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func BuildAgentMessagesFromEntries ¶
func BuildAgentMessagesFromEntries(entries []domain.ConversationEntry) []sdk.Message
BuildAgentMessagesFromEntries converts conversation entries into the flat slice of SDK messages sent to the model.
Three classes of entries are filtered:
- Plan-mode entries (entry.IsPlan): synthesized assistant messages used for UI rendering only; their content duplicates the args of the preceding RequestPlanApproval tool call.
- User-initiated bash entries: synthetic assistant + tool pairs created when the user types `!command` directly in chat. Their assistant side has tool_calls but no reasoning_content (the user, not the model, generated them).
- Pending-approval placeholders (entry.PendingToolCall != nil): UI-only empty assistant entries added while a tool awaits approval. On rejection they stay in the repo; serializing one between an assistant tool_calls message and its tool response breaks the provider's adjacency requirement ("Messages with role 'tool' must be a response to a preceding message with 'tool_calls'", issue #786). SaveConversation applies the same filter for disk persistence.
Sending the first two to a thinking-mode provider (DeepSeek, etc.) produces an assistant turn lacking `reasoning_content`, which is rejected with HTTP 400 ("The reasoning_content in the thinking mode must be passed back to the API.").
Types ¶
type Options ¶
type Options struct {
AgentService domain.AgentService
ConversationRepo domain.ConversationRepository
ModelService domain.ModelService
StateManager stateManager
Listener domain.ChatEventListener
}
Options bundles the dependencies needed to construct a Runner.
type Runner ¶
type Runner struct {
// contains filtered or unexported fields
}
func (*Runner) HandleChatChunk ¶
func (r *Runner) HandleChatChunk(msg domain.ChatChunkEvent) tea.Cmd
HandleChatChunk forwards a streaming content delta to the UI and adjusts chat status if the chunk indicates a thinking → generating transition (or vice versa).
func (*Runner) HandleChatComplete ¶
func (r *Runner) HandleChatComplete(msg domain.ChatCompleteEvent) tea.Cmd
HandleChatComplete restores the pending model (if any), updates chat status, refreshes history, emits tool-call previews, and signals completion.
func (*Runner) HandleChatError ¶
func (r *Runner) HandleChatError(msg domain.ChatErrorEvent) tea.Cmd
HandleChatError tears down session state and emits a sticky error event (with a friendlier message for "timed out" errors).
func (*Runner) HandleChatStart ¶
func (r *Runner) HandleChatStart(_ domain.ChatStartEvent) tea.Cmd
HandleChatStart transitions chat status to Starting and emits the initial "Starting response..." status. Clearing the orchestrator's active-tool indicator is the orchestrator's responsibility (see ChatHandler wrapper).
func (*Runner) HandleOptimizationStatus ¶
func (r *Runner) HandleOptimizationStatus(event domain.OptimizationStatusEvent) tea.Cmd
HandleOptimizationStatus surfaces the "Optimizing conversation..." status transitions emitted by the conversation optimizer.
func (*Runner) SetPendingRestoration ¶
SetPendingRestoration records that a temporary /model switch is in effect and that originalModel should be restored once the next completion ends.
func (*Runner) Start ¶
func (r *Runner) Start(holder domain.BashDetachChannelHolder) tea.Cmd
Start kicks off a streaming chat completion. The returned tea.Cmd performs the request (synchronously in the returned closure) and emits a ChatStartEvent on success or ChatErrorEvent on failure. The holder is attached to the request context so the agent core can find the BashDetachChannelHolder when launching tools that may need backgrounding.
Source Files