orxtra v0.13.0 /orxtra trace
On this page

Query what a run recorded: its event log, an agent session's transcript, a search across that transcript, its task tree, and its cross-agent notepad.

#orxtra trace

Query trace data: events, transcripts, tasks, and notepad entries for runs.

#trace events

Query the append-only event log the trace store keeps for one workflow run. Narrow the result with --type to a single event type such as task_started or tool_call, and cap its size with --limit, which defaults to a hundred. Honours the global --format flag, so events render as a table or as JSON.

Effect: read_only

#Flags

Flags
NameShortTypeDefaultEnvDescription
--typestrFilter events by type (e.g. task_started, tool_call).
--limitint100Maximum number of events to return from the query.

#Arguments

Arguments
NameRequiredDescription
run_idyesUnique identifier of the run to query events for (UUID format).

#trace transcript

Display the complete stored message transcript for one agent session: every message exchanged with the model, in order, as the session module persisted it. Takes the session's identifier rather than a run's. Honours the global --format flag, so the transcript renders as readable text or as JSON.

Effect: read_only

#Arguments

Arguments
NameRequiredDescription
session_idyesUnique identifier of the session to show the transcript for.

Search one agent session's stored transcript for a substring, case-insensitively, and return the matching messages rather than the whole conversation. Takes the session identifier and the text to look for. Honours the global --format flag, so matches render as a table or as JSON for a script.

Effect: read_only

#Arguments

Arguments
NameRequiredDescription
queryyesCase-insensitive substring to search for in the transcript.
session_idyesUnique identifier of the session whose transcript to search.

#trace tasks

List every task recorded for one workflow run with its current status, its attempt count and its place in the recursive task hierarchy, so you can see which branch of the tree stalled or retried. Takes the run's UUID. Honours the global --format flag, so tasks render as a table or as JSON.

Effect: read_only

#Arguments

Arguments
NameRequiredDescription
run_idyesUnique identifier of the run to list tasks for (UUID format).

#trace notepad

Show the append-only cross-agent notepad entries written during one workflow run -- the messages agents left for each other as the run progressed, in the order they were appended. Takes the run's UUID. Honours the global --format flag, so entries render as a readable table or as JSON.

Effect: read_only

#Arguments

Arguments
NameRequiredDescription
run_idyesUnique identifier of the run to show notepad entries for (UUID format).
Search