Module: Rubino::Interaction::Events

Defined in:

lib/rubino/interaction/events.rb

Overview

Defines all event types used in the system. Acts as documentation and provides constants for event names.

Constant Summary

INTERACTION_STARTED =

Interaction lifecycle events

:interaction_started

INTERACTION_FINISHED =

:interaction_finished

INTERACTION_FAILED =

:interaction_failed

INPUT_INJECTED =

Mid-turn steering: the user typed while the agent was working and the loop picked the text up at an iteration boundary, injecting it as a user message into the in-flight turn (Codex “Enter injects into the current turn”). Payload: { text:, iteration: }.

:input_injected

STATUS_CHANGED =

Status change events

:status_changed

SESSION_LOADED =

Session events

:session_loaded

SESSION_CREATED =

:session_created

SESSION_PERSISTED =

:session_persisted

MEMORY_LOADED =

Memory events

:memory_loaded

MEMORY_EXTRACTED =

:memory_extracted

MEMORY_FLUSHED =

:memory_flushed

PROMPT_ASSEMBLED =

Context events

:prompt_assembled

CONTEXT_BUDGET_CHECKED =

:context_budget_checked

COMPRESSION_STARTED =

Compression events

:compression_started

COMPRESSION_FINISHED =

:compression_finished

MODEL_CALL_STARTED =

LLM events

:model_call_started

MODEL_CALL_FINISHED =

:model_call_finished

MODEL_STREAM =

:model_stream

MESSAGE_COMPLETED =

End of one assistant message (content block). Streamed content deltas carry a message_id; this marks that block complete so a consumer can group the deltas that belong together instead of splitting them around tool calls that interleave mid-stream. Mirrors Anthropic’s content_block_stop / the AI SDK text-endid.

:message_completed

TOOL_STARTED =

Tool events

:tool_started

TOOL_PROGRESS =

Incremental progress from a long-running tool (e.g. summarize_file’s per-chunk “summarizing chunk N/M” or shell stdout lines). Emitted from the tool’s stream_chunk callback so a tool that runs for minutes without finishing keeps the API event stream alive — the SSE idle watchdog only fires when NOTHING flows, so a genuinely hung run is still caught while a busy-but-silent one heartbeats. Payload: { name:, chunk: }.

:tool_progress

TOOL_FINISHED =

:tool_finished

TOOL_APPROVAL_REQUESTED =

:tool_approval_requested

TOOL_APPROVAL_GRANTED =

:tool_approval_granted

TOOL_APPROVAL_DENIED =

:tool_approval_denied

JOB_ENQUEUED =

Job events

:job_enqueued

JOB_STARTED =

:job_started

JOB_FINISHED =

:job_finished

JOB_FAILED =

:job_failed

JOB_RETRYING =

:job_retrying

SUBAGENT_SPAWNED =

Background subagent (the ‘task` tool run in the background, the default). SPAWNED when a backgrounded subagent starts (payload: { task_id:, subagent:, prompt: }); COMPLETED/FAILED when it finishes (payload: { task_id:, subagent:, status:, output:|error: }). These let the CLI surface a completion line and the web UI show in-flight subagents —parity with how background-shell activity surfaces.

:subagent_spawned

SUBAGENT_COMPLETED =

:subagent_completed

SUBAGENT_FAILED =

:subagent_failed

SKILL_LOADED =

Skill events Emitted when the ‘skill` tool successfully loads a skill’s body into context (the level-2 “Skill ‘X’ loaded” path), so skill usage is a first-class signal for the recorder/SSE/metrics — parity with how TOOL_STARTED/SUBAGENT_* surface lifecycle. Payload: { name: } — the run association is stamped by the Recorder (run_id), like every other event.

:skill_loaded

SKILL_CREATED =

Emitted when a skill is created inline via skill(action: “create”) or by the post-turn distill job. Payload: { name:, file_path: }.

:skill_created

ARTIFACT_CREATED =

Artifact events Fired by tools that produce a downloadable user-facing file (currently AttachFileTool). Payload: { path:, filename:, content_type:, byte_size: }.

:artifact_created