Flight Recorder

POST /v1/audit/flight-recorder builds a read-only incident replay report.

It answers:


What memory could the Agent see, what was blocked, what was used, and why?

Request

The endpoint accepts either a full product_trace or individual replay artifacts.


curl -s http://127.0.0.1:3001/v1/audit/flight-recorder \
  -H 'content-type: application/json' \
  -d '{
    "tenant_id": "default",
    "scope": "checkout-migration",
    "run_id": "run-001",
    "guide_trace_id": "guide_trace:...",
    "agent_context": {},
    "memory_decision_trace": {},
    "memory_use_receipt": {},
    "operator_snapshot": {},
    "feedback_result": {},
    "decision_time": "2026-06-15T00:00:00.000Z"
  }'

At least one replay artifact is required when product_trace is absent.

SDK


import {
  agentContextFromGuide,
  memoryUseReceiptFromGuide,
} from "@aionis/sdk";
 
const guideTraceId = (guide as { guide_trace_id?: string }).guide_trace_id;
 
const report = await aionis.flightRecorder({
  run_id,
  guide_trace_id: guideTraceId,
  agent_context: agentContextFromGuide(guide),
  memory_use_receipt: memoryUseReceiptFromGuide(guide),
  feedback_result: feedback,
});

Response Surfaces

Surface	Meaning
`agent_flight_recorder.contract_version`	Stable report contract.
`agent_view.exposed_memory_ids`	Memory visible to the Agent.
`agent_view.use_now_memory_ids`	Memory admitted for direct use.
`agent_view.do_not_use_memory_ids`	Memory blocked as action guidance.
`blocked_or_suppressed`	Suppressed, stale, failed, or blocked evidence.
`attribution`	Outcome linkage to used, supported, or contradicted memory.
`claims`	Bounded audit claims the report can support.

Prompt Payload Boundary

The Flight Recorder report is designed for incident replay without prompt payload dumping. It should explain decisions, not become a second copy of every Agent prompt.

For product-level usage, see Agent Flight Recorder.