#[non_exhaustive]pub struct Item {Show 81 fields
pub item_id: String,
pub kind: ItemKind,
pub turn_id: Option<String>,
pub revision: u64,
pub status: ItemStatus,
pub recorded_at: Option<String>,
pub fallback_text: Option<String>,
pub text: Option<String>,
pub display_text: Option<String>,
pub command_id: Option<String>,
pub steered: Option<bool>,
pub retracted: Option<bool>,
pub attachments: Option<Vec<MessageAttachment>>,
pub truncated: Option<bool>,
pub message_phase: Option<MessagePhase>,
pub wake_source_descriptions: Option<Vec<String>>,
pub cut_off: Option<bool>,
pub summary: Option<Vec<String>>,
pub provider_item_id: Option<String>,
pub tool_call: ToolCallMembers,
pub command_text: Option<String>,
pub exit_code: Option<i64>,
pub exit_signal: Option<i64>,
pub duration_ms: Option<u64>,
pub subagent_id: Option<String>,
pub agent_path: Option<String>,
pub depth: Option<u32>,
pub role: Option<String>,
pub objective: Option<String>,
pub child_session_id: Option<String>,
pub control_status: Option<SubagentControlStatus>,
pub workflow_run_id: Option<String>,
pub usage: Option<TokenUsage>,
pub result: Option<SubagentResult>,
pub history_unreadable: Option<bool>,
pub entry_id: Option<String>,
pub script_id: Option<String>,
pub trigger_source: Option<String>,
pub resume_from_run_id: Option<String>,
pub paused: Option<bool>,
pub children: Option<Vec<WorkflowChild>>,
pub message: Option<String>,
pub token_budget: Option<WorkflowTokenBudget>,
pub declared_phase: Option<String>,
pub final_summary: Option<String>,
pub stop_available: Option<bool>,
pub child_prompt_refs: Option<Vec<OutputRef>>,
pub reminder_agent_id: Option<String>,
pub generation_id: Option<u64>,
pub task_id: Option<String>,
pub child_session_log_path: Option<String>,
pub blocking: Option<bool>,
pub side_session_id: Option<String>,
pub selector: Option<String>,
pub source: Option<String>,
pub outcome: Option<CompactionOutcome>,
pub trigger: Option<CompactionTrigger>,
pub strategy_id: Option<String>,
pub summarized_through: Option<String>,
pub tokens_before: Option<u64>,
pub tokens_after: Option<u64>,
pub reason: Option<String>,
pub hook_key: Option<String>,
pub event: Option<HookEventName>,
pub run_status: Option<HookRunStatus>,
pub label: Option<String>,
pub action: Option<String>,
pub query: Option<String>,
pub url: Option<String>,
pub pattern: Option<String>,
pub provider_status: Option<String>,
pub results: Option<Vec<WebSearchHit>>,
pub raw_item: Option<String>,
pub monitor_id: Option<String>,
pub monitor_description: Option<String>,
pub damaged_range: Option<DamagedRange>,
pub background_terminal_completion: Option<BackgroundTerminalCompletion>,
pub earlier_history: Option<EarlierHistory>,
pub scheduled_fire: ScheduledFireMembers,
pub terminal_row_after_item_id: Option<String>,
pub replay_hint: Option<ReplayHint>,
}Expand description
One transcript item at one revision. item/started, item/updated, and
item/completed carry the full object; item/delta appends to it by field
path.
Fields (Non-exhaustive)§
This struct is marked as non-exhaustive
Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.item_id: StringBare UUIDv7; the item’s identity across its whole lifecycle. Identity
rule: the task id for task-backed kinds (toolCall, subagent), the
pre-minted commit message_id for delta-streamed kinds (agentMessage,
reasoning), EventId::derive over the session id and the range’s
starting sequence number for damagedRange, the same over the boundary
checkpoint’s sequence number for earlierHistory, and the opening
durable record’s event id for everything else. Opaque to clients.
kind: ItemKindThe item kind (open enum): clients MUST render unknown kinds generically
— kind name plus status plus fallbackText.
turn_id: Option<String>The owning turn. null for the kinds outside a turn: userShell
always, hookRun for session-boundary runs, monitorFeed always,
sideChat always, damagedRange always, earlierHistory always.
revision: u64Integer >= 1, strictly monotonic per item; apply rule is
replace-iff-higher. item/delta never bumps it.
status: ItemStatusOpen enum; terminal = anything other than "inProgress". Unknown values
MUST be treated as terminal-unknown and rendered generically.
recorded_at: Option<String>RFC3339 timestamp of the item’s driving durable record. Absent on ephemeral-opened items until first durable re-emission.
fallback_text: Option<String>Server-provided one-line human summary any kind MAY carry, for generic rendering of kinds a client does not recognize; new kinds SHOULD carry it during their first release cycle.
text: Option<String>userMessage: the prompt text as submitted. agentMessage: the
accumulated reply, streamed via item/delta field "text".
reasoning: raw committed reasoning text where the provider exposes it.
display_text: Option<String>userMessage: presentation form; absent when the client sent none.
command_id: Option<String>userMessage/userShell: the submitting command — multi-client UIs
de-duplicate their local echo on it.
steered: Option<bool>userMessage: true when injected mid-turn via turn/steer or
ifBusy: "steer"; absent otherwise.
retracted: Option<bool>userMessage: true after an accepted retract; re-emitted via
item/updated.
attachments: Option<Vec<MessageAttachment>>userMessage: image attachment metadata only — base64 payloads are not
echoed back on the view.
truncated: Option<bool>true when the server’s per-surface text budget saturated a streamed
surface (agentMessage.text, reasoning.summary[*],
toolCall.visibleOutput, userShell.visibleOutput), cut a subagent
surface (objective, fallbackText, result.summary, result.text),
or cut a monitorFeed surface (text, fallbackText,
monitorDescription). On a subagent item the flag is item-level and
never cleared: once any of those four is cut, every later update keeps
it, even when result.text arrives whole. The durable full text remains
in the log.
message_phase: Option<MessagePhase>agentMessage: the phase the model declared on the commit; see MessagePhase.
wake_source_descriptions: Option<Vec<String>>agentMessage: the Monitor source descriptions for the wake turn’s
first answer — stripped, bounded, one entry per arming drain in drain
order. Absent (or empty) means untagged: the row renders the ordinary
untagged shape. The client scrubs per part before joining.
cut_off: Option<bool>agentMessage: true when the provider stream died mid-reply and the
run failed with this partial text committed — the wire mark for the
gate-off “incomplete response” rule. Absent otherwise; never cleared
once set.
summary: Option<Vec<String>>reasoning: one entry per summary part; part n streams via
item/delta field "summary.n" (part boundary = index change).
provider_item_id: Option<String>reasoning: provider reasoning item id (e.g. rs_...), for
provider-side correlation.
tool_call: ToolCallMemberstoolCall members, flattened beside the other optional members.
command_text: Option<String>userShell: the command text as submitted.
exit_code: Option<i64>userShell: the process exit code, when it exited by code.
exit_signal: Option<i64>userShell: the terminating signal NUMBER, when signalled (e.g. 9) —
the durable payload verbatim; nothing maps numbers to names, and the
fold never invents one.
duration_ms: Option<u64>userShell/subagent: observed wall-clock duration. workflow: the
run’s wall time from launch to reconciliation, set only on the settling
item/completed — never a sum of child durations. toolCall: the
Backgrounded row’s duration — the open→update record delta in whole ms,
served on the backgroundStatus re-emission and frozen there; absent
otherwise.
subagent_id: Option<String>subagent: durable child identity (subagent_id).
agent_path: Option<String>subagent: agent definition path.
depth: Option<u32>subagent: nesting depth.
role: Option<String>subagent: role as spawned.
objective: Option<String>subagent: objective as spawned, bounded on the wire at the 64 KiB
retained-text budget; a cut sets truncated. The durable record stays
verbatim.
child_session_id: Option<String>subagent/reminderChild: the child’s own session id, readable via
session/read/view/page — child transcript drill-down without a
second protocol.
control_status: Option<SubagentControlStatus>subagent: camelCased SubagentControlStatus (open enum); status
stays the generic item vocabulary.
workflow_run_id: Option<String>subagent/workflow: the owning durable workflow run id (opaque
string — not a UUID family).
usage: Option<TokenUsage>subagent: transitive observed usage — the child and its own
descendants; updated in place, absent until first observation; never
folded into session/tokenUsage.cumulative.
result: Option<SubagentResult>subagent: the result envelope from ResultReady.
history_unreadable: Option<bool>subagent: the host could not read this live child’s run records when
the parent resumed; a client says so once. A resume-time observation:
present only as true, absent otherwise, never false, and never
carried by a live notification.
entry_id: Option<String>workflow: launched entry identity.
script_id: Option<String>workflow: launched script identity.
trigger_source: Option<String>workflow: camelCased WorkflowLaunchTriggerSource, verbatim
(durable runtime vocabulary, e.g. "modelProposal").
resume_from_run_id: Option<String>workflow: set on resumed launches.
paused: Option<bool>workflow: true only while a durable pause fact is in force for the
run’s generation; absent otherwise, never false.
children: Option<Vec<WorkflowChild>>workflow: folded per-child state, keyed by (childId, attempt),
re-emitted whole on every change — the item revision is the ordering
guard.
message: Option<String>workflow: the reconciled terminal message, set on completion.
token_budget: Option<WorkflowTokenBudget>workflow: the run’s launch token budget, set from the launch fact;
absent when the launch carried none.
declared_phase: Option<String>workflow: the run-level declared-phase label — the last Phase
detail-row label of the reconciled projection; terminal-only, absent
while live.
final_summary: Option<String>workflow: the human terminal summary projected from the reconciled
Launch envelope, bounded at the 64 KiB surface budget; terminal-only.
stop_available: Option<bool>workflow: display-only stop affordance — true while the run is open,
false once settled; stopping rides workflow/cancel. Authorizes
nothing.
child_prompt_refs: Option<Vec<OutputRef>>workflow: one outputRef-shaped link per recorded child spawn with an
objective (kind: "workflow_child_prompt"), fetched via
item/readOutput.
reminder_agent_id: Option<String>reminderChild: the reminder agent’s id.
generation_id: Option<u64>reminderChild: the reminder generation.
task_id: Option<String>reminderChild: the linked task id.
child_session_log_path: Option<String>reminderChild: parent-session-dir-relative child log path; absent
when no filesystem log.
blocking: Option<bool>reminderChild: true when this child is a blocking reminder (it holds
the turn open until it settles); absent otherwise, never false. The
status word stays off the wire; clients name the reminder from
reminderAgentId.
side_session_id: Option<String>sideChat: the side’s own session id, readable via
session/read/view/page — drill-down without a second protocol.
selector: Option<String>skillRead: the read skill’s catalog selector (matches the skill/list
row; the raw observed id when the read resolved to no catalog row). Set
on every skillRead item; old clients render it beside the kind name.
source: Option<String>skillRead: the catalog source (matches the skill/list row); empty
when the host could not resolve the read to a catalog row (replay
without the tool result, context-only evidence). Set on every
skillRead item; each client composes the notice line from selector +
source.
outcome: Option<CompactionOutcome>compaction: terminal only — folds installed/fallback status.
trigger: Option<CompactionTrigger>compaction: what initiated the compaction.
strategy_id: Option<String>compaction: summarizer strategy (installed only).
summarized_through: Option<String>compaction: opaque provenance string naming the compaction boundary.
Also a compaction-anchored read anchor, accepted by session/resume and
view/page. Still opaque — clients relay it, never parse it.
tokens_before: Option<u64>compaction: token budget snapshot before, when measured.
tokens_after: Option<u64>compaction: token budget snapshot after, when measured.
reason: Option<String>compaction: noop/failure reason, verbatim (snake_case durable
vocabulary, e.g. "no_compactable_history"). hookRun: terminal-only
capped terminal reason.
hook_key: Option<String>hookRun: the durable hook_key.
event: Option<HookEventName>hookRun: the hook event in Q2 camelCase vocabulary.
run_status: Option<HookRunStatus>hookRun: terminal-only durable outcome; stays blocked when the item
settles rejected.
label: Option<String>hookRun: capped display label.
action: Option<String>webSearch: the search action as an open-enum string — search |
openPage | findInPage, else tolerated unknown. Absent while the call
is open (the action is not final until the item completes).
query: Option<String>webSearch: the search query, verbatim — present iff action == "search" with a query to carry.
url: Option<String>webSearch: the page URL, verbatim — present iff action == "openPage"
or "findInPage".
pattern: Option<String>webSearch: the in-page pattern, verbatim — present iff action == "findInPage".
provider_status: Option<String>webSearch: the provider’s terminal status, verbatim ("completed" |
"failed" | "cancelled" | other) — present on the completed item.
results: Option<Vec<WebSearchHit>>webSearch: the typed result hits, verbatim, never truncated — clients
read results.length for the count.
raw_item: Option<String>webSearch: the verbatim raw provider JSON frame under the 64 KiB
per-surface budget; past-budget frames cut here and ride whole behind
outputRef.
monitor_id: Option<String>monitorFeed: the monitor task id — the same value as that monitor’s
toolCall itemId.
monitor_description: Option<String>monitorFeed: the monitor’s arm-time description with the durable
"Monitor event: " summary prefix stripped; model- controlled tool
input under the 64 KiB per-surface budget — sanitize before render.
damaged_range: Option<DamagedRange>damagedRange: what the damaged range cost.
background_terminal_completion: Option<BackgroundTerminalCompletion>backgroundTerminalCompletion: the five facts the Finished cell
renders.
earlier_history: Option<EarlierHistory>earlierHistory: the boundary checkpoint’s transcript tail.
scheduled_fire: ScheduledFireMembersscheduledFire members, flattened beside the other optional members.
terminal_row_after_item_id: Option<String>Terminal toolCall: the item after whose rows the terminal row renders.
replay_hint: Option<ReplayHint>Monitor toolCall: the hint verdict plus anchor (ReplayHint); absent
means no hint.