first commit

src/api/utils/emulator_response_builder.py

@@ -0,0 +1,202 @@
+"""Builds the `TicketResponseEvent` for the Response Emulator flow.
+
+Kept separate from `agent_helpers.build_cms_response_event` (checklist's
+builder) to avoid coupling the two domains. Status mapping (status,action):
+    error at any step                           → failed, retry
+    flow_mode="generate"                        → awaiting_review, await_response
+    flow_mode="approve"                         → approved, await_response
+    flow_mode="close"                           → done, response
+
+The event is meant to be the LAST message published on the OCI Response
+Stream for a given case, so the CMS callback resolves the final DB state
+without races against earlier ProgressEvents.
+
+## Where each field lives (split by lifetime)
+
+- `processing.metadata` (per-run scratch, lives only in the latest
+  `processing` subdoc): just `error` when the graph failed. Everything else
+  used to be here too and was duplicating the case-doc `metadata` below —
+  the GET endpoint already reads validation/selected_actions from the
+  case-level `metadata`, so keeping a parallel copy under `processing` was
+  pure dead weight that the CMS overwrote on every run.
+- `metadata` (top-level case doc, persisted incrementally via `$set` with
+  dot-notation): selected_actions, validation, flow_mode, emulator_routing,
+  and the `last_emulation` summary. These survive across runs and merge
+  cleanly with whatever the checklist already wrote to `metadata`.
+"""
+
+from datetime import datetime, timezone
+
+from src.agent.state.agent_state import AgentState
+from src.agent.state.steps_emulator import EmulatorGraphStep
+from src.api.schemas.anatel_schemas import (
+    Processing,
+    ProcessingAction,
+    ProcessingStatus,
+    TicketResponseEvent,
+)
+
+
+# Maps flow_mode → (status, action) when the graph completes without error.
+_FLOW_MODE_OUTCOMES: dict[str, tuple[ProcessingStatus, ProcessingAction]] = {
+    "generate": (ProcessingStatus.AWAITING_REVIEW, ProcessingAction.AWAIT_RESPONSE),
+    "approve": (ProcessingStatus.APPROVED, ProcessingAction.AWAIT_RESPONSE),
+    "close": (ProcessingStatus.DONE, ProcessingAction.RESPONSE),
+}
+
+
+def _resolve_crm_protocol(case_data: dict | None) -> str | None:
+    """Returns the persisted Siebel SR protocol, or None when no SR exists.
+
+    Why this is restricted to two sources (and NOT `case_data.crmProtocol`
+    at the root or `request_context.crmProtocol`): the root-level field on
+    the request event is whatever the simulator/CMS sent in — for our test
+    payloads it's `"DS-XXXXXXXX"`, a ticketId-shaped placeholder, not the
+    real Anatel-style SR protocol (`"20260..."`) that Siebel returns when
+    the treatment SR opens. Falling back to it would propagate that bogus
+    value into `processing.crmProtocol` and the close PATCH would close
+    the wrong SR.
+
+    Trustworthy sources, in order:
+      1. `processing.crmProtocol` — written by the checklist's terminal
+         event from `siebel_sr_data.interactionProtocol`. Canonical.
+      2. `siebel_sr_data.interactionProtocol` — in-memory only; relevant
+         when the checklist + emulator run in the same process (tests).
+
+    The emulator MUST echo this back on every terminal event because the
+    CMS callback overwrites the `processing` subdoc; omitting the field
+    zeroes it, which is the bug that broke `close_case` before.
+    """
+    case_data = case_data or {}
+    processing = case_data.get("processing") or {}
+    sr_data = case_data.get("siebel_sr_data") or {}
+    return processing.get("crmProtocol") or sr_data.get("interactionProtocol")
+
+
+def _resolve_persisted_field(
+    case_data: dict | None,
+    new_value,
+    field: str,
+):
+    """Returns `new_value` when truthy, else the persisted value in `case_data.processing.{field}`.
+
+    Why: when a graph run fails (e.g. `approve_draft` errors because of a
+    transient issue), the state's `metadata.{field}` is empty for that
+    run. If we publish `Processing(field=None)`, the CMS callback wipes
+    the value the previous successful run had persisted — turning a
+    recoverable failure into permanent data loss (this is exactly what
+    nuked `case_response`/`transitions` mid-flow before this fix). When
+    the new run actually produced a value, that wins; otherwise we echo
+    what is already in the doc so the failure is non-destructive.
+    """
+    if new_value:
+        return new_value
+    processing = (case_data or {}).get("processing") or {}
+    return processing.get(field)
+
+
+def _resolve_outcome(
+    error_info: dict | None,
+    case_response: str | None,
+    flow_mode: str | None,
+) -> tuple[ProcessingStatus, ProcessingAction]:
+    if error_info and flow_mode == "close":
+        return ProcessingStatus.SIEBEL_CLOSING_FAILED, ProcessingAction.RETRY
+    if error_info:
+        return ProcessingStatus.FAILED, ProcessingAction.RETRY
+    if flow_mode == "close" and not case_response:
+        return ProcessingStatus.FAILED, ProcessingAction.RETRY
+    if flow_mode == "generate" and not case_response:
+        return ProcessingStatus.FAILED, ProcessingAction.RETRY
+    return _FLOW_MODE_OUTCOMES.get(
+        flow_mode or "",
+        (ProcessingStatus.FAILED, ProcessingAction.RETRY),
+    )
+
+
+def _build_case_metadata(
+    selected_actions: list,
+    validation: dict,
+    flow_mode: str | None,
+    routing: dict | None,
+    emulation_type: str | None,
+    iteration_count: int,
+) -> dict:
+    """Top-level case-doc metadata for the CMS to `$set` via dot notation.
+
+    Empty/None values are omitted so a partial run (e.g. `approve` has no
+    new `selected_actions` or `validation`) doesn't blow away keys written
+    by an earlier run — this is the incremental-merge contract with the
+    CMS callback. `last_emulation` is always included as a fresh marker
+    of when the most recent graph run happened and what it was.
+    """
+    case_metadata: dict = {
+        "last_emulation": {
+            "type": emulation_type,
+            "is_regeneration": emulation_type == "regenerate",
+            "flow_mode": flow_mode,
+            "iteration_count": iteration_count,
+            "at": datetime.now(timezone.utc).isoformat(),
+        },
+    }
+    if selected_actions:
+        case_metadata["selected_actions"] = selected_actions
+    if validation:
+        case_metadata["validation"] = validation
+    if flow_mode:
+        case_metadata["flow_mode"] = flow_mode
+    if routing:
+        case_metadata["emulator_routing"] = routing
+    return case_metadata
+
+
+def build_emulator_response_event(state: AgentState, transaction_id: str) -> TicketResponseEvent:
+    metadata = state.get("metadata", {}) or {}
+    request_context = metadata.get("request_context") or {}
+    case_response = metadata.get("case_response") or request_context.get("case_response")
+    current_step = str(state.get("current_step") or "")
+    error_info = state.get("error")
+    validation = metadata.get("validation") or {}
+    flow_mode = metadata.get("flow_mode") or request_context.get("flow_mode")
+    transitions = metadata.get("transitions")
+    selected_actions = metadata.get("selected_actions") or []
+    routing = metadata.get("emulator_routing")
+    emulation_type = request_context.get("type")
+    iteration_count = state.get("iteration_count", 0)
+    case_data = metadata.get("case_data") or {}
+    crm_protocol = _resolve_crm_protocol(case_data)
+    # Echo persisted values when this run didn't produce new ones so a
+    # failed run doesn't blank out the previous successful draft.
+    case_response = _resolve_persisted_field(case_data, case_response, "case_response")
+    transitions = _resolve_persisted_field(case_data, transitions, "transitions")
+
+    cms_status, action = _resolve_outcome(error_info, case_response, flow_mode)
+
+    # Per-run scratch only — case-level info goes under top-level `metadata`.
+    response_metadata: dict = {}
+    if error_info:
+        response_metadata["error"] = error_info
+
+    case_metadata = _build_case_metadata(
+        selected_actions=selected_actions,
+        validation=validation,
+        flow_mode=flow_mode,
+        routing=routing,
+        emulation_type=emulation_type,
+        iteration_count=iteration_count,
+    )
+
+    return TicketResponseEvent(
+        transactionId=transaction_id,
+        processing=Processing(
+            status=cms_status,
+            current_step=current_step or str(EmulatorGraphStep.CASE_CLOSED),
+            action=action,
+            note=state.get("processing_notes") or None,
+            crmProtocol=crm_protocol,
+            case_response=case_response,
+            transitions=transitions,
+            metadata=response_metadata or None,
+        ),
+        metadata=case_metadata,
+    )