From dbb94a31ee8fbde486ffef3f9e3d15a3296a56be Mon Sep 17 00:00:00 2001
From: Seongho Bae <seongho@kw.ac.kr>
Date: Sat, 11 Jul 2026 15:18:08 +0900
Subject: [PATCH] fix(opencode): retry invalid control output

---
 .github/workflows/opencode-review.yml        |  8 +++++---
 scripts/ci/run_opencode_review_model_pool.sh | 18 ++++++++++--------
 tests/test_opencode_agent_contract.py        |  5 +++--
 3 files changed, 18 insertions(+), 13 deletions(-)

diff --git a/.github/workflows/opencode-review.yml b/.github/workflows/opencode-review.yml
index 3bfe0db2c..9e195c2d1 100644
--- a/.github/workflows/opencode-review.yml
+++ b/.github/workflows/opencode-review.yml
@@ -2975,9 +2975,11 @@ jobs:
           OPENCODE_RUN_TIMEOUT_SECONDS: "180"
           OPENCODE_EXPORT_TIMEOUT_SECONDS: "60"
           OPENCODE_TOTAL_RETRY_BUDGET_SECONDS: "540"
-          # Stop after one catalog pass; the retry budget should fail closed
-          # with visible diagnostics before the 350-min job timeout.
-          OPENCODE_POOL_MAX_CYCLES: "1"
+          # Keep cycling through the high-sensitivity catalog until a model
+          # emits a valid control block or the bounded retry/step timeout fires.
+          # A single invalid no-control response is provider/tool state, not a
+          # source finding.
+          OPENCODE_POOL_CYCLE_SLEEP_SECONDS: "15"
           OPENCODE_BACKOFF_INITIAL_SECONDS: "5"
           OPENCODE_BACKOFF_MAX_SECONDS: "5"
           OPENCODE_FIRST_ATTEMPT_AGENT: ci-review
diff --git a/scripts/ci/run_opencode_review_model_pool.sh b/scripts/ci/run_opencode_review_model_pool.sh
index 44ef11a93..cd3686854 100644
--- a/scripts/ci/run_opencode_review_model_pool.sh
+++ b/scripts/ci/run_opencode_review_model_pool.sh
@@ -278,13 +278,15 @@ main() {
 				if [ "$run_status" -eq 2 ]; then
 					break
 				fi
-				retry_sleep="$(backoff_sleep "$attempt")"
-				if [ "$deadline" -gt 0 ] && [ $((SECONDS + retry_sleep)) -gt "$deadline" ]; then
-					retry_sleep=$((deadline - SECONDS))
-				fi
-				if [ "$retry_sleep" -gt 0 ]; then
-					printf 'Retrying OpenCode after exponential backoff of %ss.\n' "$retry_sleep"
-					sleep "$retry_sleep"
+				if [ "$attempt" -lt "$attempts" ]; then
+					retry_sleep="$(backoff_sleep "$attempt")"
+					if [ "$deadline" -gt 0 ] && [ $((SECONDS + retry_sleep)) -gt "$deadline" ]; then
+						retry_sleep=$((deadline - SECONDS))
+					fi
+					if [ "$retry_sleep" -gt 0 ]; then
+						printf 'Retrying OpenCode after exponential backoff of %ss.\n' "$retry_sleep"
+						sleep "$retry_sleep"
+					fi
 				fi
 			done
 		done
@@ -295,7 +297,7 @@ main() {
 			record_pool_exhausted
 			exit 1
 		fi
-		printf 'OpenCode retry budget and the workflow step timeout remain the outer guards for provider stalls.\n'
+		printf 'OpenCode retry budget and the workflow step timeout remain the outer guards for invalid or unavailable provider output.\n'
 		cycle_sleep="${OPENCODE_POOL_CYCLE_SLEEP_SECONDS:-60}"
 		if [ "$deadline" -gt 0 ] && [ $((SECONDS + cycle_sleep)) -gt "$deadline" ]; then
 			cycle_sleep=$((deadline - SECONDS))
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index f548d828e..368275858 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -446,7 +446,8 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     assert 'OPENCODE_RUN_TIMEOUT_SECONDS: "180"' in workflow
     assert 'OPENCODE_EXPORT_TIMEOUT_SECONDS: "60"' in workflow
     assert 'OPENCODE_TOTAL_RETRY_BUDGET_SECONDS: "540"' in workflow
-    assert 'OPENCODE_POOL_MAX_CYCLES: "1"' in workflow
+    assert "OPENCODE_POOL_MAX_CYCLES" not in workflow
+    assert 'OPENCODE_POOL_CYCLE_SLEEP_SECONDS: "15"' in workflow
     assert 'OPENCODE_BACKOFF_MAX_SECONDS: "5"' in workflow
     assert 'OPENCODE_EXHAUSTED_REKICK_INITIAL_SLEEP_SECONDS: "15"' in workflow
     assert 'OPENCODE_EXHAUSTED_REKICK_MAX_SLEEP_SECONDS: "30"' in workflow
@@ -462,7 +463,7 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     assert "OpenCode model pool has no configured model candidates." in model_pool_runner
     assert 'OPENCODE_TOTAL_RETRY_BUDGET_SECONDS:-2400' in model_pool_runner
     assert "completed a full model-candidate cycle without a valid control conclusion" in model_pool_runner
-    assert "retry budget and the workflow step timeout" in model_pool_runner
+    assert "workflow step timeout remain the outer guards for invalid or unavailable provider output" in model_pool_runner
     assert "OpenCode model pool exhausted before producing a valid control conclusion." in model_pool_runner
     assert 'record_review_status "exhausted"' in model_pool_runner
     assert "retry budget exhausted" not in model_pool_runner