fix(autoresearch): preserve typed representation verdicts

Keep host-generated representation continuations substantive enough for the supervisor report gate so successful state transitions are not mislabeled as evaluation failures.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: 2 people

scripts/agent_gan_repl.py

@@ -7946,7 +7946,10 @@ def build_autoresearch_verdict(
             )
         )
         frontier = (
-            target.statement if target is not None else
+            (
+                f"Continue unresolved target {target.obligation_id}: "
+                f"{target.statement}"
+            ) if target is not None else
             "Construct a concrete smaller proof obligation."
         )
     return {

tests/inference_engine/bridge/test_agent_gan_repl.py

@@ -4315,12 +4315,16 @@ class Candidate:
     assert verdict["outcome"] == "DECOMPOSED"
     assert verdict["created_obligation_ids"] == ["RH-C2-child"]
     ledger.obligations[1].decomposition_certificate_hash = ""
-    assert build_autoresearch_verdict(
+    inconclusive = build_autoresearch_verdict(
         Candidate,
         ledger,
         {"RH-C2": "UNRESOLVED"},
         [ledger.obligations[1]],
-    )["outcome"] == "INCONCLUSIVE"
+    )
+    assert inconclusive["outcome"] == "INCONCLUSIVE"
+    assert inconclusive["new_frontier"] == (
+        "Continue unresolved target RH-C2: Prove zero convergence."
+    )
 
 
 def test_critic_leaf_table_cannot_bypass_certificate():