StreamState

When True, stream-time diagnostics (the tool-call and file-operation notifications, plus the stdout separator newline preceding a tool call) are suppressed, so stdout carries only agent response text.

When True (default), text chunks are written to stdout as they arrive.

When False, text is buffered in full_response and flushed after the agent finishes.

Accumulated text fragments from the AI message stream.

Maps a tool-call index or ID to its in-progress buffer: name, ID, accumulated argument fragments, and the display latch.

Maps in-flight tool-call IDs (those whose tool.use has fired) to their name and parsed arguments, so the matching tool.result can be correlated. Entries are removed when the result arrives; any still present when the stream aborts are closed by _dispatch_orphaned_tool_result_hooks. One record per id keeps name and args structurally in lock-step (see InFlightToolCall).

Tool-call IDs whose non-interactive call line has already been printed.

Tool-call IDs for which a tool.use has been dispatched.

Monotonic: never cleared within a run, so tool.use fires at most once per id even if a call's arg chunks are redelivered after its result (mirrors the TUI's displayed_tool_ids). Result correlation and the orphan drain use the separate in_flight_tool_calls, which is cleared per result.

Maps interrupt IDs to their validated HITL requests that are awaiting decisions.

Maps interrupt IDs to dicts containing a 'decisions' key with a list of decision dicts (each having a 'type' key of 'approve' or 'reject').

Used to resume the agent after HITL processing.

Raw Hooks v2 invocation interrupt payloads awaiting client fulfillment.

Resume values for fulfilled Hooks v2 interrupts, keyed by interrupt id.

Client-side Hooks v2 coordinator for this headless session.

Completed root and identified subagent stream messages.

Model projected into compact lifecycle events.

Whether the current stream crossed a compaction boundary.

Compaction tool results whose post-boundary lifecycle already fired.

Whether the headless client session has emitted its terminal event.

Flag indicating whether any HITL interrupt was received during the current stream pass.

Accumulated model usage stats for this stream.

Requests already counted in this headless run, keyed by message ID.

Monotonic across HITL resume passes so a replayed message does not add its request, tokens, or cost to stats again. Each pass closes its entries via finalize_recorded_requests, which is what extends that guarantee to replayed chunks -- an open chunked request accepts revisions, so without the round boundary a replayed chunk would merge into it a second time.

Optional animated spinner shown during agent work in verbose mode.

Whether rubric lifecycle messages should include iteration numbers.