Gets the cross-config sandbox impact of this execution.
Returns one row per integration config the run touched in sandbox mode, with the pre-run snapshot id (if the session was auto-snapshotted), the diff against current state (added/modified/removed entities), and the simulator tool-call count. Backs the "Sandbox impact" card in AgentLab and on the execution detail page. Vendor-routed configs do not appear here — they are surfaced on the frontend from the agent revision's bound configs.
Authorization
Bearer JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.
In: header
Path Parameters
The agent identifier.
The execution identifier.
The tenant identifier
Response Body
curl -X GET "https://api.fruxon.com/v1/tenants/string/agents/string/executionRecords/string/sandboxImpact"{
"executionId": "00000000-0000-0000-0000-000000000000",
"runGroupId": "00000000-0000-0000-0000-000000000000",
"configs": [
{
"integrationConfigId": "00000000-0000-0000-0000-000000000000",
"integrationId": "string",
"displayName": "string",
"mode": "NONE",
"preRunSnapshotId": "00000000-0000-0000-0000-000000000000",
"diff": {
"added": [],
"modified": [],
"removed": []
},
"callCount": 0
}
]
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}Removes all evaluation samples derived from this execution
Deletes every sample row across every dataset that was sourced from this execution; use this to fully retract a run from evaluation tooling when it was tagged in error or contains data that shouldn't be retained. Only the dataset cross-references are removed — the underlying execution record, its trace, and any cost data are kept intact. Returns 204 even when there were no samples to delete.
Lists execution sessions for an agent — one item per spawn tree, newest activity first.
The grouped counterpart of the flat list: each item is a whole session (a root run plus every sub-task and consult descending from it) collapsed to its root, carrying the same rollups the `/tree` endpoint computes — run count, failures, combined cost, wall-clock span. A session of one is just a run that spawned nothing. Filters apply at the run level and lift to the session: a session is returned when ANY of its runs matches, and `matchedRuns` reports how many did — so filtering by a tool error surfaces the whole session containing the hit, with its root for context. Rollups always cover the whole session regardless of the filter. Pagination counts sessions, not runs.