fix(runtime-e2e): verify unknown default + drop stale merge-status note (#2912)

runtime-e2e/caller_name_audit/README.md

@@ -20,6 +20,11 @@ that it marshals onto the wire correctly (that's covered by the
    equals that value verbatim.
 2. Legacy `tool_type="mcp"` with **no** `caller_name` → the deprecated
    fallback still resolves `policy_details.caller_name` to `"mcp"`.
+3. **Neither** `caller_name` **nor** `tool_type` supplied → the platform's
+   default-fallback (getaxonflow/axonflow-enterprise#2903, folded into the
+   same #2953 merge) resolves `policy_details.caller_name` to `"unknown"` —
+   not the pre-#2903 default of `"claude_code"`. An unidentified caller must
+   never be silently attributed to a specific client.
 
 The SDK's typed `AuditLogEntry` does not declare a `policy_details` field
 (it's an internal JSONB blob), so the read side uses a raw `httpx` call
@@ -39,18 +44,17 @@ local-dev Enterprise stack in `axonflow-enterprise/main-tree/.env`). In
 Community mode, tenant-audit reads are tenant-wide regardless, so the header
 is a no-op there.
 
-## Prerequisite: platform support is not yet on `main`
+## Prerequisite: platform version
 
-`caller_name` support (axonflow-enterprise#2953) is implemented but, as of
-this writing, still an open PR on the `feat/2912-caller-name-tool-type-deprecation`
-branch — not yet merged to `axonflow-enterprise` main. Against a stack built
-from `axonflow-enterprise` main, this test will FAIL (the 45s poll of
-`GET /api/v1/audit/tenant/{tenant_id}` times out waiting for
-`policy_details.caller_name`, which the server doesn't write yet) — that's
-not a bug in this test, it means the platform side isn't deployed on
-whatever stack you're pointed at. Point your local `axonflow-enterprise`
-checkout at that branch (or a later commit that includes it) before running
-this test.
+`caller_name` support (axonflow-enterprise#2953, which also folded in the
+#2903 default-fallback fix) is merged to `axonflow-enterprise` main and
+shipped in platform release **v9.11.0**. Point your stack at v9.11.0+ (or a
+`main` checkout at/after commit `7a5984ec7`) before running this test.
+Against an older platform, the third scenario (`unknown` default) and
+possibly `policy_details.caller_name` itself will not resolve as expected,
+and the 45s poll of `GET /api/v1/audit/tenant/{tenant_id}` will time out
+waiting for a row that never lands with the expected shape — that's a
+platform-version mismatch, not a bug in this test.
 
 ## Run
 
@@ -65,7 +69,8 @@ The orchestrator's `AuditLogger` batches writes (flush every 10s per
 `platform/orchestrator/audit_logger.go`), so the test polls
 `GET /api/v1/audit/tenant/{tenant_id}` for up to 45s before failing.
 
-Exits non-zero if `caller_name` (or the `tool_type` fallback) does not reach
+Exits non-zero if `caller_name`, the `tool_type` fallback, or the
+neither-supplied `"unknown"` default does not reach
 `policy_details.caller_name` on the real row.
 
 ## Companion unit coverage

runtime-e2e/caller_name_audit/test.py

@@ -22,10 +22,16 @@
   2. Call it again with only the legacy `tool_type` set (no `caller_name`)
      to prove the backward-compat fallback still resolves into
      `policy_details.caller_name` on a real row.
-  3. The orchestrator's AuditLogger batches writes (flush every 10s), so
+  3. Call it a third time with NEITHER `caller_name` NOR `tool_type` set,
+     to prove the platform's default-fallback (getaxonflow/axonflow-
+     enterprise#2903, folded into the same #2953 merge) resolves to
+     `"unknown"` -- an unidentified caller must NOT be silently attributed
+     to a specific client. `"unknown"` replaced the pre-#2903 default of
+     `"claude_code"`.
+  4. The orchestrator's AuditLogger batches writes (flush every 10s), so
      poll `GET /api/v1/audit/tenant/{tenant_id}` (the same route
      `client.get_audit_logs_by_tenant` hits) until each row lands.
-  4. The SDK's typed `AuditLogEntry` does not surface `policy_details`
+  5. The SDK's typed `AuditLogEntry` does not surface `policy_details`
      (it's an internal JSONB blob), so this step reads the raw JSON body
      directly -- the only way to observe `policy_details.caller_name`
      from outside the platform -- and asserts it equals what we sent.
@@ -77,6 +83,8 @@
 WORKFLOW_CALLER_NAME = f"e2e-caller-name-wf-{_RUN_ID}"
 WORKFLOW_TOOL_TYPE_FALLBACK = f"e2e-tool-type-fallback-wf-{_RUN_ID}"
 WANT_TOOL_TYPE_FALLBACK = "mcp"
+WORKFLOW_NEITHER_SUPPLIED = f"e2e-neither-supplied-wf-{_RUN_ID}"
+WANT_DEFAULT_CALLER_NAME = "unknown"  # #2903: was "claude_code" pre-fix
 
 # Batch writer flushes every 10s (platform/orchestrator/audit_logger.go);
 # give it generous headroom under CI/local load.
@@ -168,6 +176,19 @@ async def main() -> None:
             f"SDK audit_tool_call (tool_type fallback) -> audit_id={resp2.audit_id} status={resp2.status}"
         )
 
+        # 3. Neither caller_name nor tool_type supplied -- the platform's
+        # default-fallback (#2903) must resolve to "unknown", never silently
+        # attribute an unidentified caller to a specific client.
+        resp3 = await client.audit_tool_call(
+            AuditToolCallRequest(
+                tool_name="e2eNeitherSuppliedTool",
+                workflow_id=WORKFLOW_NEITHER_SUPPLIED,
+            )
+        )
+        print(
+            f"SDK audit_tool_call (neither supplied) -> audit_id={resp3.audit_id} status={resp3.status}"
+        )
+
     policy_details_1 = _fetch_policy_details(WORKFLOW_CALLER_NAME)
     got_caller_name = policy_details_1.get("caller_name")
     if got_caller_name != WANT_CALLER_NAME:
@@ -195,6 +216,19 @@ async def main() -> None:
     )
     print(f"Wire policy_details (tool_type fallback case): {policy_details_2}")
 
+    policy_details_3 = _fetch_policy_details(WORKFLOW_NEITHER_SUPPLIED)
+    got_default = policy_details_3.get("caller_name")
+    if got_default != WANT_DEFAULT_CALLER_NAME:
+        _fail(
+            f"default fallback: policy_details.caller_name = {got_default!r}, want "
+            f"{WANT_DEFAULT_CALLER_NAME!r} (#2903) (full policy_details: {policy_details_3})"
+        )
+    print(
+        f"PASS: neither caller_name nor tool_type supplied -> "
+        f"policy_details.caller_name = {got_default!r} (#2903 default, not 'claude_code')"
+    )
+    print(f"Wire policy_details (neither-supplied case): {policy_details_3}")
+
     print("ALL PASS: caller_name (#2912) verified end-to-end through the real SDK + platform")