Kiro Glossary

What Kiro’s V3 agent engine calls the things the project glossary names, and where each one appears in the files it writes.

The dialect is kiro/1. That names the vocabulary, not the adapter: kiro-local says how records were acquired, the dialect says how they are read.

Two words here mean something other than what a reader might expect. Kiro’s execution is one turn, which is a Run here. And Kiro writes a subagent’s records in a file of their own, a sub-execution, which is a child stream here.

The tables below are copied from what asz glossary kiro/1 prints. The command reads the adapter’s own declarations, so where this page and the command differ, the command is right.

asz glossary kiro/1                 this table
asz conversation ID                 talk · run · tool
asz conversation -terms native ID   talk · executionId · tool_call and tool_result
asz conversation -terms both ID     run  (executionId)

61 terms: 24 that Kiro names, 37 this project derives.

Named by Kiro

Each row says where the thing is recorded.

Model Kiro Where Note
agent.call sub_agent_start messages.jsonl no tool call names the spawn; the record is the call
agent.output sub_agent_complete messages.jsonl the response the parent received; the child’s own last record is in its file
child subSessionId sub_agent_start and sub_agent_complete also the name of the child’s file under sub-executions
context.injection agent_note, session_start, steering_inclusion messages.jsonl and a user record with syntheticUserMessageReason, the engine’s text around a steering message
control.notice turn_end, ContextualHookInvoked messages.jsonl a turn_end whose stopReason is not end_turn; also a record with no type, or a type this adapter does not describe
conversation id session.json one session is one conversation; Kiro Crew names no chain between the sessions of one chat
epoch.boundary tombstone messages.jsonl kind summarization or checkpoint_revert
epoch.summary assistant, operationType Summary messages.jsonl nothing names the tombstone it follows
error.api session_metadata, key displayError messages.jsonl
external user messages.jsonl
id id every record followed by # and the first 12 hex digits of the record’s SHA-256; the engine gives one id to an approval it asks twice
message.assistant assistant, operationType Say messages.jsonl
message.external user messages.jsonl also a message sent during a turn, source steer; Kiro Crew writes its own prompts here too
redacted reasoningRedactedContent a Reasoning record
result_of toolCallId tool_result subSessionId for a subagent’s end
run executionId turn_start and the records of the turn a prompt carries none; it belongs to the first turn named after it in the file
session session sessions//sess_
starts subSessionId sub_agent_start
stream sub-execution sub-executions/.jsonl the main stream is messages.jsonl
thinking assistant, operationType Reasoning messages.jsonl the text … means the engine kept no reasoning
time timestamp every record not in line order; line order is the order
tool tool_call and tool_result messages.jsonl and sub-executions joined by toolCallId
turn.duration usage_summary messages.jsonl elapsedTime, with the turn’s credits; no token counts
unavailable ... the content of a Reasoning record

Derived by this project

Nothing in the files carries these. They are what assembly works out from the evidence above, and saying so is more useful than a plausible name that would send a reader looking for a field that is not there.

Model Kiro Where Note
agent.launch_ack — derived here no equivalent: nothing acknowledges a subagent separately
available — derived here the engine keeps text and tool results whole, apart from large tool output
batch — derived here no equivalent: subagents started together share only their turn
call — derived here no equivalent: a turn lists its request ids in usage_summary, and no record names one
cancels — derived here no equivalent
conflict — derived here derived
continues — derived here no equivalent: a tombstone’s effectiveFromMessageId names where the dropped records begin, not the last one kept
control.command — derived here no equivalent: a slash command leaves no record of its own
control.interrupt — derived here not read: a turn the person cancelled ends with stopReason cancelled, which lands as a notice
control.permission — derived here not read: pending_interaction and interaction_resolved are landed and are no step
ends_with — derived here derived from a child’s last response
epoch — derived here the span between two tombstones
exact_ambiguous — derived here derived
exact_unique — derived here one toolCallId or subSessionId on both records
follows — derived here derived from line order
hash_only — derived here no equivalent
in_segment — derived here derived
input_of — derived here no equivalent
llm.call — derived here no equivalent: the engine records what the model said and not which request said it
message.synthetic — derived here not read: records the engine writes on recovery end their id in -synthetic, are a tool result, a child’s end or a turn’s end, and none was measured
model — derived here not read: no record says which model answered; session.json names the model chosen
notification — derived here no equivalent
omitted — derived here no equivalent
parent — derived here no equivalent: a record names no record before it
reports — derived here no equivalent
retries — derived here no equivalent
runtime.notification — derived here no equivalent: a subagent’s end is the result of its call
segment — derived here an activity window
size_only — derived here no equivalent
strong_inference — derived here derived
summarizes — derived here not read: nothing joins a summary to its tombstone
talk — derived here one readable interaction: a prompt and the turn it opened
trigger — derived here derived: a user record that is neither a steering note nor written by the engine
truncated — derived here not read: the model is sent a preview of a large tool output, and that preview is the result here; the whole output is in tool-outputs, which is not collected
unknown — derived here derived
unresolved — derived here derived
weak_inference — derived here derived