Activity

Activities represent individual events in an agent session’s conversation history. An Activity is a union type – exactly one of the following payload fields will be set.

Field	Type	Description
user	UserActivity	A message or action from a non-LLM actor (human user, system, or tool results)
assistant	AssistantActivity	A message or action generated by the LLM
error	Error	An error that occurred during execution
compaction	Compaction	A summarization of earlier conversation history
slack_thread	SlackThreadActivity	A reply in a Slack thread the agent is participating in
slack_mention	SlackMentionActivity	An @-mention of the agent in Slack
slack_reaction	SlackReactionActivity	An emoji reaction added to an agent-sent message in Slack
external_agent	ExternalAgentActivity	A message from another agent
task_transition	TaskTransition	A boundary between tasks: a subtask call, a tail call, or a resume after the session waited for the next message
artifacts_changed	ArtifactsChanged	The session’s artifact set changed (user-initiated PUT/DELETE through the session artifacts endpoint)
timestamp	timestamp	When this activity occurred

User Activity

A message or action from a non-LLM actor.

Field	Type	Description
text	Text	Text content of the user message
tool_results	ToolResult[]	Results from tool calls the user (or system) executed
author	string	Author identifier

Assistant Activity

A message or action generated by the LLM.

Field	Type	Description
text	Text	Generated text content
tool_calls	ToolCall[]	Tool calls the assistant wants to execute
usage	Usage	Model identity and token accounting for the LLM call that produced this turn. Unset for turns that did not originate from a model call.

Slack Thread Activity

A reply posted to a Slack thread the agent is participating in.

Field	Type	Description
permalink	string	Full Slack URL to the thread
user_id	string	Slack user ID of the person who replied
text	string	Reply content
channel_name	string	Name of the Slack channel

Slack Mention Activity

An @Firetiger mention in a Slack message.

Field	Type	Description
permalink	string	Slack URL to the mention
user_id	string	Slack user ID of the person who mentioned the agent
text	string	Message text (with @mention stripped)
channel_id	string	Slack channel ID (server-resolved from channel_name when seeded via InvokeTrigger; never caller-supplied on the public surface)
channel_name	string	Name of the Slack channel
thread_ts	string	Thread timestamp
message_ts	string	Message timestamp
idempotency_key	string	Stamped by TriggersService.InvokeTrigger from InvokeTriggerRequest.idempotency_key. The Slack event dispatcher sets this to a per-(event, trigger) key so retries of the same Slack event return the prior session instead of creating a duplicate. Empty when InvokeTrigger is called without an idempotency key.

Slack Reaction Activity

An emoji reaction added to an agent-sent message in Slack. Used to capture user feedback on agent responses.

Field	Type	Description
permalink	string	Slack URL to the reacted message
user_id	string	Slack user ID of the person who added the reaction
reaction	string	Emoji name without colons (e.g., “thumbsup”, “+1”, “white_check_mark”)
channel_id	string	Slack channel ID
channel_name	string	Name of the Slack channel
message_ts	string	Timestamp of the message that was reacted to
thread_ts	string	Thread timestamp if the reacted message is in a thread

External Agent Activity

A message from another agent session.

Field	Type	Description
source_session	string	Resource name of the source agent session
source_agent_name	string	Human-readable name of the source agent
title	string	Message title
summary	string	Brief summary
content	string	Full content (markdown supported)

Task Transition

A boundary between tasks within an agent session. The runtime writes one when control moves from one task to another. Clients reading session history can use kind to interpret why the task changed.

Field	Type	Description
from_task	string	Name of the task control moved from
to_task	string	Name of the task control moved to
kind	TaskTransitionKind	Why the transition happened (see below)
input	object	Structured input for the new task; resolves ${field} references in its prompt
context	TaskTransitionContext	How the new task receives prior conversation context (defaults to hidden — no prior context)

kind is one of:

Value	Description
TASK_TRANSITION_KIND_CALL	A subtask was spawned; the parent suspends until it returns
TASK_TRANSITION_KIND_TAIL_CALL	A tail call into another task of the same agent (the state graph)
TASK_TRANSITION_KIND_AGENT_TAIL_CALL	A tail call into another agent
TASK_TRANSITION_KIND_END_TURN_RESUME	The session ended its turn, waited, and resumed into the declared task when the next message arrived — a user reply in interactive mode, or a peer/scheduler push in autonomous mode
TASK_TRANSITION_KIND_ASK_RESUME	Deprecated. Collapsed into TASK_TRANSITION_KIND_END_TURN_RESUME; retained only so sessions written before the migration remain readable
TASK_TRANSITION_KIND_PAUSE_RESUME	Deprecated. Collapsed into TASK_TRANSITION_KIND_END_TURN_RESUME; retained only so sessions written before the migration remain readable

Artifacts Changed

A signal-only event indicating that the session’s artifact set under the artifacts/ namespace has changed — a user-initiated PUT or DELETE through the session artifacts endpoint completed successfully. Carries no payload: consumers re-derive truth by listing the session’s artifacts. Tool-result writeArtifact calls do not emit this event; their effect is already visible via the surrounding ToolResult’s artifacts.

Compaction

A summarization that replaces earlier conversation history to keep sessions manageable.

Field	Type	Description
summary	string	Summary of the preceding messages that were compacted
truncated_token_count	integer	Number of tokens that were truncated

Text

Field	Type	Description
content	string	The text content
role	MessageRole	The sender role: USER, ASSISTANT, or SYSTEM. Set USER for a client-supplied seed message.

Usage

Model identity and token accounting for a single LLM call, mirroring the OpenTelemetry gen_ai semantic conventions. Carried on AssistantActivity so observability backends can attribute and price each turn.

Field	Type	Description
model	string	Request model id, e.g. claude-opus-4-5-20251101
provider	string	Serving platform: anthropic, aws.bedrock, gcp.vertex_ai, openai, or baseten
input_tokens	int64	Prompt tokens consumed
output_tokens	int64	Completion tokens generated
cache_read_tokens	int64	Prompt-cache hit tokens (0 if no cache hit)
cache_write_tokens	int64	Prompt-cache creation tokens (0 if no cache write)
finish_reason	string	Provider stop reason, e.g. end_turn, tool_use, max_tokens

Tool Call

A request from the assistant to invoke a tool.

Field	Type	Description
id	string	Unique identifier for this tool call (referenced by the corresponding ToolResult)
name	string	Name of the tool to invoke (e.g. TOOL_POSTGRES_QUERY)
arguments	string	JSON-encoded arguments for the tool

Tool Result

The result of executing a tool call.

Field	Type	Description
tool_call_id	string	ID of the ToolCall this result corresponds to
content	string	Output of the tool execution
is_error	boolean	Whether the tool execution failed

Error

An error that occurred during session execution.

Field	Type	Description
message	string	Human-readable error message