From 4076f39c7f0668a67088e7d94fa1fd6470e0804b Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Thu, 30 Jul 2026 17:22:01 +0900
Subject: [PATCH 01/10] feat(review): add NVIDIA NIM provider

---
 .../workflows/opencode-review-dispatch.yml    | 26 +++++++++++++---
 scripts/ci/run_opencode_review_model_pool.sh  | 13 +++++++-
 scripts/ci/test_strix_quick_gate.sh           |  6 ++++
 tests/test_opencode_agent_contract.py         |  9 ++++++
 tests/test_opencode_model_pool_runner.py      | 31 +++++++++++++++++++
 5 files changed, 80 insertions(+), 5 deletions(-)

diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index 1966a7562..ee5555a70 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -3341,7 +3341,7 @@ jobs:
             "$schema": "https://opencode.ai/config.json",
             "model": "github-models/deepseek/deepseek-r1-0528",
             "small_model": "github-models/deepseek/deepseek-v3-0324",
-            "enabled_providers": ["opencode-free", "openai", "openrouter", "github-models"],
+            "enabled_providers": ["opencode-free", "nvidia-nim", "openai", "openrouter", "github-models"],
             "lsp": false,
             "mcp": {},
             "permission": {
@@ -3528,6 +3528,24 @@ jobs:
                   }
                 }
               },
+              "nvidia-nim": {
+                "npm": "@ai-sdk/openai-compatible",
+                "name": "NVIDIA NIM",
+                "options": {
+                  "baseURL": "https://integrate.api.nvidia.com/v1",
+                  "apiKey": "{env:NVIDIA_NIM_API_KEY}"
+                },
+                "models": {
+                  "nvidia/nemotron-3-ultra-550b-a55b": {
+                    "name": "NVIDIA Nemotron 3 Ultra 550B A55B",
+                    "tool_call": true,
+                    "limit": {
+                      "context": 1000000,
+                      "output": 32768
+                    }
+                  }
+                }
+              },
               "openai": {
                 "npm": "@ai-sdk/openai",
                 "name": "OpenAI (direct)",
@@ -3838,6 +3856,7 @@ jobs:
           # in the opencode.jsonc "openai" provider block.
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+          NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
           SHARE: "false"
           NPM_CONFIG_IGNORE_SCRIPTS: "true"
           NO_COLOR: "1"
@@ -3846,8 +3865,7 @@ jobs:
           # Their trial/free-period data
           # may be logged, retained, or used for product/model improvement, so
           # private repositories never include them. The pool then falls
-          # through to
-          # DeepSeek V3, the direct GPT-5.6 Luna slot, pinned PAID
+          # through to NVIDIA NIM, DeepSeek V3, the direct GPT-5.6 Luna slot, pinned PAID
           # OpenRouter coder models (free-tier candidates hit the shared
           # free-models-per-day cap and hung for the full candidate timeout,
           # so the OpenRouter slots use cheap paid models billed against the
@@ -3857,7 +3875,7 @@ jobs:
           # cost-efficient tier, cheaper than the legacy gpt-5 it replaced
           # ($1/$6 vs $1.25/$10 per 1M tokens) so the org OpenAI budget
           # stretches further between top-ups.
-          OPENCODE_MODEL_CANDIDATES: "${{ needs.validate-pr-metadata.outputs.is_private == 'false' && 'opencode-free/nemotron-3-ultra-free opencode-free/deepseek-v4-flash-free opencode-free/north-mini-code-free opencode-free/laguna-s-2.1-free opencode-free/ling-3.0-flash-free opencode-free/big-pickle opencode-free/mimo-v2.5-free ' || '' }}github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5 github-models/openai/gpt-5-chat github-models/openai/o3 github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1"
+          OPENCODE_MODEL_CANDIDATES: "${{ needs.validate-pr-metadata.outputs.is_private == 'false' && 'opencode-free/nemotron-3-ultra-free opencode-free/deepseek-v4-flash-free opencode-free/north-mini-code-free opencode-free/laguna-s-2.1-free opencode-free/ling-3.0-flash-free opencode-free/big-pickle opencode-free/mimo-v2.5-free ' || '' }}nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5 github-models/openai/gpt-5-chat github-models/openai/o3 github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1"
           # One attempt per model, then fall through to the next model. Retrying
           # the SAME model 5x let a rate-limited/hung leader consume the whole
           # step, so the pool never reached a healthy fallback model.
diff --git a/scripts/ci/run_opencode_review_model_pool.sh b/scripts/ci/run_opencode_review_model_pool.sh
index 3c4583069..fcb7ae84e 100644
--- a/scripts/ci/run_opencode_review_model_pool.sh
+++ b/scripts/ci/run_opencode_review_model_pool.sh
@@ -345,6 +345,13 @@ is_openrouter_candidate() {
 	esac
 }
 
+is_nvidia_nim_candidate() {
+	case "$1" in
+	nvidia-nim/*) return 0 ;;
+	*) return 1 ;;
+	esac
+}
+
 is_low_sensitivity_candidate() {
 	case "$1" in
 	openai/*-mini | openai/*-nano | \
@@ -372,6 +379,10 @@ should_skip_model_candidate() {
 		printf 'Skipping OpenCode %s because OPENROUTER_API_KEY is not configured; falling back to the next provider-qualified candidate.\n' "$model_candidate"
 		return 0
 	fi
+	if is_nvidia_nim_candidate "$model_candidate" && [ -z "${NVIDIA_NIM_API_KEY:-}" ]; then
+		printf 'Skipping OpenCode %s because NVIDIA_NIM_API_KEY is not configured; falling back to the next provider-qualified candidate.\n' "$model_candidate"
+		return 0
+	fi
 	return 1
 }
 
@@ -381,7 +392,7 @@ cap_model_run_timeout() {
 	local cap_seconds
 
 	case "$model_candidate" in
-	opencode-free/*)
+	opencode-free/* | nvidia-nim/*)
 		cap_seconds="$(env_integer_or_default OPENCODE_FREE_RUN_TIMEOUT_SECONDS 600)"
 		;;
 	github-models/openai/gpt-5 | github-models/openai/gpt-5-chat)
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index c05353570..01ab19476 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -599,6 +599,10 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$workflow_file" 'is_private: ${{ steps.validate.outputs.is_private }}' "opencode review carries validated repository privacy into model routing"
 	assert_file_contains "$workflow_file" '"opencode-free"' "opencode review enables its anonymous Zen free provider"
 	assert_file_contains "$workflow_file" '"baseURL": "https://opencode.ai/zen/v1"' "opencode review routes the free provider through the official Zen endpoint"
+	assert_file_contains "$workflow_file" '"nvidia-nim"' "opencode review enables its NVIDIA NIM provider"
+	assert_file_contains "$workflow_file" '"baseURL": "https://integrate.api.nvidia.com/v1"' "opencode review routes NVIDIA NIM through its official hosted endpoint"
+	assert_file_contains "$workflow_file" '"apiKey": "{env:NVIDIA_NIM_API_KEY}"' "opencode review resolves NVIDIA NIM credentials at runtime"
+	assert_file_contains "$workflow_file" 'NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}' "opencode review exposes NVIDIA NIM credentials only to the model runtime"
 	assert_file_contains "$workflow_file" '"north-mini-code-free"' "opencode review declares the current Zen coding model"
 	assert_file_contains "$workflow_file" "needs.validate-pr-metadata.outputs.is_private == 'false'" "opencode review limits data-retaining free models to public repositories"
 	assert_file_matches "$workflow_file" 'uses:[[:space:]]+actions/checkout@[0-9a-fA-F]{40}([[:space:]]|$)' "opencode review workflow pins checkout to a full commit SHA"
@@ -860,6 +864,7 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OpenCode model pool has no configured model candidates." "opencode model pool fails fast when no candidates are configured"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OPENAI_API_KEY is not configured" "opencode model pool skips native OpenAI candidates when the org secret is absent"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OPENROUTER_API_KEY is not configured" "opencode model pool skips OpenRouter candidates when the org secret is absent"
+	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "NVIDIA_NIM_API_KEY is not configured" "opencode model pool skips NVIDIA NIM candidates when the org secret is absent"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "configured max cycle count" "opencode model pool exits before the job timeout after configured cycles"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_TOTAL_RETRY_BUDGET_SECONDS:-1500' "opencode model pool keeps a bounded default retry budget unless the workflow explicitly disables it"
 	assert_file_not_contains "$workflow_file" "no model produced a valid review control block" "opencode model-failure path no longer documents a final exhausted state"
@@ -1197,6 +1202,7 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$workflow_file" 'GH_TOKEN: ${{ secrets.OPENCODE_APPROVE_TOKEN || steps.review_read_app_token.outputs.token || github.token }}' "opencode manual dispatch uses the cross-repo approval token for target PR evidence lookups with app-token fallback"
 	assert_file_contains "$workflow_file" 'repos/${GH_REPOSITORY}' "opencode review workflow uses env-backed repository context in shell commands"
 	assert_file_contains "$workflow_file" "Run OpenCode PR Review model pool" "opencode review starts the central model pool"
+	assert_file_contains "$workflow_file" "nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b" "opencode review tries the current NVIDIA NIM Nemotron model"
 	assert_file_contains "$workflow_file" "github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5" "opencode review starts with DeepSeek V3 before full-size GPT fallbacks"
 	assert_file_contains "$workflow_file" "github-models/deepseek/deepseek-r1-0528" "opencode review keeps a reachable DeepSeek R1 reasoning fallback model"
 	assert_file_contains "$workflow_file" "github-models/deepseek/deepseek-v3-0324" "opencode review has a reachable DeepSeek V3 fallback model"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index bd300a839..ccf946a74 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -134,6 +134,7 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
         ["opencode-free", "ling-3.0-flash-free"],
         ["opencode-free", "big-pickle"],
         ["opencode-free", "mimo-v2.5-free"],
+        ["nvidia-nim", "nvidia/nemotron-3-ultra-550b-a55b"],
         ["github-models", "deepseek/deepseek-v3-0324"],
         ["openai", "gpt-5.6-luna"],
         ["openrouter", "deepseek/deepseek-v3.2"],
@@ -160,6 +161,14 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
     )
     assert generated_config_match is not None
     generated_config = json.loads(generated_config_match.group(1))
+    nvidia_provider = generated_config["provider"]["nvidia-nim"]
+    assert nvidia_provider["options"] == {
+        "baseURL": "https://integrate.api.nvidia.com/v1",
+        "apiKey": "{env:NVIDIA_NIM_API_KEY}",
+    }
+    assert nvidia_provider["models"]["nvidia/nemotron-3-ultra-550b-a55b"][
+        "limit"
+    ] == {"context": 1000000, "output": 32768}
     free_models = generated_config["provider"]["opencode-free"]["models"]
     assert set(free_models) == {
         "nemotron-3-ultra-free",
diff --git a/tests/test_opencode_model_pool_runner.py b/tests/test_opencode_model_pool_runner.py
index 3fd432242..e4949e67a 100644
--- a/tests/test_opencode_model_pool_runner.py
+++ b/tests/test_opencode_model_pool_runner.py
@@ -790,6 +790,37 @@ def test_free_provider_runtime_cap_preserves_queue_budget(tmp_path: Path) -> Non
     ) in result.stdout
 
 
+def test_nvidia_nim_candidate_requires_key(tmp_path: Path) -> None:
+    """NVIDIA NIM is skipped cleanly when its scoped credential is unavailable."""
+    result = run_failed_model(
+        tmp_path,
+        model_candidates="nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
+    )
+
+    assert result.returncode == 1
+    assert "NVIDIA_NIM_API_KEY is not configured" in result.stdout
+    assert "attempt 1/1" not in result.stdout
+
+
+def test_nvidia_nim_runtime_cap_preserves_queue_budget(tmp_path: Path) -> None:
+    """A stalled hosted NIM cannot consume a full paid-provider cadence slot."""
+    result = run_failed_model(
+        tmp_path,
+        extra_env={
+            "NVIDIA_NIM_API_KEY": "fake-nvidia-key",
+            "OPENCODE_FREE_RUN_TIMEOUT_SECONDS": "3",
+            "OPENCODE_RUN_TIMEOUT_SECONDS": "9",
+        },
+        model_candidates="nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
+    )
+
+    assert result.returncode == 1
+    assert (
+        "OpenCode nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b runtime cap "
+        "selected 3s instead of 9s because this provider has a bounded failover window."
+    ) in result.stdout
+
+
 def test_github_models_openai_prompt_references_evidence_without_inlining(
     tmp_path: Path,
 ) -> None:

From bea81270085fa06fddfbd9b930c9cda1164773e9 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Thu, 30 Jul 2026 17:47:38 +0900
Subject: [PATCH 02/10] feat(review): route public scanners through NVIDIA NIM

---
 .github/workflows/noema-review.yml            | 24 +++++++
 .../workflows/opencode-review-dispatch.yml    |  2 +-
 .github/workflows/strix.yml                   | 70 ++++++++++++++++---
 ...opencode_failed_check_fallback_findings.sh |  6 +-
 scripts/ci/test_strix_quick_gate.sh           | 13 +++-
 tests/test_opencode_agent_contract.py         | 11 ++-
 .../test_required_workflow_queue_contract.py  |  5 ++
 7 files changed, 113 insertions(+), 18 deletions(-)

diff --git a/.github/workflows/noema-review.yml b/.github/workflows/noema-review.yml
index e52371d43..dd6156274 100644
--- a/.github/workflows/noema-review.yml
+++ b/.github/workflows/noema-review.yml
@@ -250,6 +250,23 @@ jobs:
           echo "::add-mask::$app_token"
           echo "token=$app_token" >>"$GITHUB_OUTPUT"
 
+      - name: Resolve Noema target repository visibility
+        if: env.PR_NUMBER != ''
+        id: target_visibility
+        env:
+          GH_TOKEN: ${{ secrets.NOEMA_REVIEW_TOKEN || steps.noema_github_app_token.outputs.token || steps.noema_oidc_token.outputs.token }}
+        run: |
+          set -euo pipefail
+          is_private="$(gh api "repos/${TARGET_REPOSITORY}" --jq '.private')"
+          case "$is_private" in
+            true | false) ;;
+            *)
+              echo "::error::Noema target repository visibility did not resolve to true or false."
+              exit 1
+              ;;
+          esac
+          echo "is_private=$is_private" >>"$GITHUB_OUTPUT"
+
       - name: Run Noema LLM review and submit verdict
         if: env.PR_NUMBER != ''
         env:
@@ -258,6 +275,8 @@ jobs:
           NOEMA_LLM_API_URL: ${{ vars.NOEMA_LLM_API_URL || '' }}
           NOEMA_LLM_MODEL: ${{ vars.NOEMA_LLM_MODEL || '' }}
           NOEMA_LLM_API_KEY: ${{ secrets.NOEMA_LLM_API_KEY || secrets.OPENAI_API_KEY || '' }}
+          NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
+          TARGET_REPOSITORY_PRIVATE: ${{ steps.target_visibility.outputs.is_private }}
         run: |
           set -euo pipefail
           if [ -z "${PR_NUMBER:-}" ]; then
@@ -268,6 +287,11 @@ jobs:
             echo "::error::Noema reviewer credential selection succeeded but no token was minted; review cannot submit a verdict."
             exit 1
           fi
+          if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ] && [ -z "${NOEMA_LLM_API_URL:-}" ] && [ -z "${NOEMA_LLM_MODEL:-}" ]; then
+            export NOEMA_LLM_API_URL="https://integrate.api.nvidia.com/v1/chat/completions"
+            export NOEMA_LLM_MODEL="nvidia/nemotron-3-ultra-550b-a55b"
+            export NOEMA_LLM_API_KEY="${NVIDIA_NIM_API_KEY:-}"
+          fi
           if [ -z "${NOEMA_LLM_API_URL:-}" ] || [ -z "${NOEMA_LLM_MODEL:-}" ] || [ -z "${NOEMA_LLM_API_KEY:-}" ]; then
             echo "::error::Noema LLM is unconfigured: NOEMA_LLM_API_URL, NOEMA_LLM_MODEL, and NOEMA_LLM_API_KEY (or OPENAI_API_KEY) are required."
             exit 1
diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index ee5555a70..2ec38f6b8 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -3875,7 +3875,7 @@ jobs:
           # cost-efficient tier, cheaper than the legacy gpt-5 it replaced
           # ($1/$6 vs $1.25/$10 per 1M tokens) so the org OpenAI budget
           # stretches further between top-ups.
-          OPENCODE_MODEL_CANDIDATES: "${{ needs.validate-pr-metadata.outputs.is_private == 'false' && 'opencode-free/nemotron-3-ultra-free opencode-free/deepseek-v4-flash-free opencode-free/north-mini-code-free opencode-free/laguna-s-2.1-free opencode-free/ling-3.0-flash-free opencode-free/big-pickle opencode-free/mimo-v2.5-free ' || '' }}nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5 github-models/openai/gpt-5-chat github-models/openai/o3 github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1"
+          OPENCODE_MODEL_CANDIDATES: "${{ needs.validate-pr-metadata.outputs.is_private == 'false' && 'opencode-free/nemotron-3-ultra-free opencode-free/deepseek-v4-flash-free opencode-free/north-mini-code-free opencode-free/laguna-s-2.1-free opencode-free/ling-3.0-flash-free opencode-free/big-pickle opencode-free/mimo-v2.5-free nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b ' || '' }}github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5 github-models/openai/gpt-5-chat github-models/openai/o3 github-models/deepseek/deepseek-r1-0528 github-models/deepseek/deepseek-r1"
           # One attempt per model, then fall through to the next model. Retrying
           # the SAME model 5x let a rate-limited/hung leader consume the whole
           # step, so the pool never reached a healthy fallback model.
diff --git a/.github/workflows/strix.yml b/.github/workflows/strix.yml
index afc040246..4f95a31f6 100644
--- a/.github/workflows/strix.yml
+++ b/.github/workflows/strix.yml
@@ -258,6 +258,27 @@ jobs:
             echo "token=$app_token"
           } >>"$GITHUB_OUTPUT"
 
+      - name: Resolve target repository visibility
+        id: target_visibility
+        env:
+          GH_TOKEN: ${{ steps.target_app_token.outputs.token || secrets.OPENCODE_APPROVE_TOKEN || github.token }}
+          TARGET_REPOSITORY: ${{ github.event.client_payload.target_repository || github.event.pull_request.base.repo.full_name || github.repository }}
+        run: |
+          set -euo pipefail
+          if [[ ! "$TARGET_REPOSITORY" =~ ^ContextualWisdomLab/[A-Za-z0-9_.-]+$ ]]; then
+            echo "::error::Strix target repository must belong to ContextualWisdomLab."
+            exit 1
+          fi
+          is_private="$(gh api "repos/${TARGET_REPOSITORY}" --jq '.private')"
+          case "$is_private" in
+            true | false) ;;
+            *)
+              echo "::error::Target repository visibility did not resolve to true or false."
+              exit 1
+              ;;
+          esac
+          echo "is_private=$is_private" >>"$GITHUB_OUTPUT"
+
       - name: Materialize target workspace
         if: github.event_name != 'repository_dispatch'
         env:
@@ -422,11 +443,13 @@ jobs:
       - name: Gate Strix secrets
         id: gate
         env:
-          STRIX_MODEL: ${{ github.event.client_payload.strix_llm || 'gpt-5.6-luna' }}
+          STRIX_MODEL: ${{ github.event.client_payload.strix_llm || (steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna') }}
           STRIX_OPENAI_API_KEY: ${{ secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY }}
           STRIX_OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+          STRIX_NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
           STRIX_VERTEX_CREDENTIALS: ${{ secrets.GCP_SA_KEY }}
           STRIX_GITHUB_MODELS_TOKEN: ${{ secrets.STRIX_GITHUB_MODELS_TOKEN || github.token }}
+          TARGET_REPOSITORY_PRIVATE: ${{ steps.target_visibility.outputs.is_private }}
         run: |
           strix_model="$(printf '%s' "$STRIX_MODEL" | sed 's/^[[:space:]]*//;s/[[:space:]]*$//')"
           case "$strix_model" in
@@ -469,6 +492,20 @@ jobs:
                 exit 1
               fi
               ;;
+            nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b)
+              if [ "$TARGET_REPOSITORY_PRIVATE" != "false" ]; then
+                echo '::error::NVIDIA NIM hosted trial scans are limited to public repositories.'
+                exit 1
+              fi
+              echo 'enabled=true' >> "$GITHUB_OUTPUT"
+              echo 'provider_mode=nvidia_nim' >> "$GITHUB_OUTPUT"
+              sanitized_nvidia_key="$(printf '%s' "$STRIX_NVIDIA_NIM_API_KEY" | tr -d '\r\n')"
+              trimmed_nvidia_key="$(printf '%s' "$sanitized_nvidia_key" | sed 's/^[[:space:]]*//;s/[[:space:]]*$//')"
+              if [ -z "$trimmed_nvidia_key" ]; then
+                echo '::error::NVIDIA_NIM_API_KEY is required for Strix NVIDIA NIM scans.'
+                exit 1
+              fi
+              ;;
             vertex_ai/gemini-3.1-pro-preview-customtools | vertex_ai/gemini-2.5-flash)
               echo 'enabled=true' >> "$GITHUB_OUTPUT"
               echo 'provider_mode=vertex_ai' >> "$GITHUB_OUTPUT"
@@ -480,7 +517,7 @@ jobs:
               fi
               ;;
             *)
-              echo '::error::STRIX_LLM must select GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model.'
+              echo '::error::STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model.'
               exit 1
               ;;
           esac
@@ -550,7 +587,7 @@ jobs:
       - name: Mask LLM API key
         if: steps.gate.outputs.enabled == 'true'
         env:
-          LLM_API_KEY: ${{ steps.gate.outputs.provider_mode == 'github_models' && (secrets.STRIX_GITHUB_MODELS_TOKEN || github.token) || steps.gate.outputs.provider_mode == 'openai_direct' && (secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY) || steps.gate.outputs.provider_mode == 'openrouter' && secrets.OPENROUTER_API_KEY || '' }}
+          LLM_API_KEY: ${{ steps.gate.outputs.provider_mode == 'github_models' && (secrets.STRIX_GITHUB_MODELS_TOKEN || github.token) || steps.gate.outputs.provider_mode == 'openai_direct' && (secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY) || steps.gate.outputs.provider_mode == 'openrouter' && secrets.OPENROUTER_API_KEY || steps.gate.outputs.provider_mode == 'nvidia_nim' && secrets.NVIDIA_NIM_API_KEY || '' }}
         run: |
           # Sanitize CR/LF before masking to prevent broken ::add-mask::
           # commands and potential workflow command injection.
@@ -566,7 +603,7 @@ jobs:
       - name: Prepare LLM API key input file
         if: steps.gate.outputs.enabled == 'true'
         env:
-          LLM_API_KEY_SECRET: ${{ steps.gate.outputs.provider_mode == 'github_models' && (secrets.STRIX_GITHUB_MODELS_TOKEN || github.token) || steps.gate.outputs.provider_mode == 'openai_direct' && (secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY) || steps.gate.outputs.provider_mode == 'openrouter' && secrets.OPENROUTER_API_KEY || '' }}
+          LLM_API_KEY_SECRET: ${{ steps.gate.outputs.provider_mode == 'github_models' && (secrets.STRIX_GITHUB_MODELS_TOKEN || github.token) || steps.gate.outputs.provider_mode == 'openai_direct' && (secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY) || steps.gate.outputs.provider_mode == 'openrouter' && secrets.OPENROUTER_API_KEY || steps.gate.outputs.provider_mode == 'nvidia_nim' && secrets.NVIDIA_NIM_API_KEY || '' }}
           PROVIDER_MODE: ${{ steps.gate.outputs.provider_mode }}
         run: |
           sanitized="$(printf '%s' "$LLM_API_KEY_SECRET" | tr -d '\r\n')"
@@ -583,6 +620,10 @@ jobs:
             echo '::error::OPENROUTER_API_KEY is required for Strix OpenRouter scans.'
             exit 1
           fi
+          if [ -z "$trimmed" ] && [ "$PROVIDER_MODE" = "nvidia_nim" ]; then
+            echo '::error::NVIDIA_NIM_API_KEY is required for Strix NVIDIA NIM scans.'
+            exit 1
+          fi
           umask 077
           llm_api_key_file="$RUNNER_TEMP/llm_api_key.txt"
           printf '%s' "$trimmed" > "$llm_api_key_file"
@@ -596,6 +637,14 @@ jobs:
           printf '%s' 'https://openrouter.ai/api/v1' > "$llm_api_base_file"
           echo "LLM_API_BASE_FILE=$llm_api_base_file" >> "$GITHUB_ENV"
 
+      - name: Prepare NVIDIA NIM API base
+        if: steps.gate.outputs.provider_mode == 'nvidia_nim'
+        run: |
+          umask 077
+          llm_api_base_file="$RUNNER_TEMP/llm_api_base.txt"
+          printf '%s' 'https://integrate.api.nvidia.com/v1' > "$llm_api_base_file"
+          echo "LLM_API_BASE_FILE=$llm_api_base_file" >> "$GITHUB_ENV"
+
       - name: Prepare GitHub Models API base
         if: steps.gate.outputs.provider_mode == 'github_models'
         run: |
@@ -605,7 +654,7 @@ jobs:
           echo "LLM_API_BASE_FILE=$llm_api_base_file" >> "$GITHUB_ENV"
 
       - name: Prepare GitHub Models fallback credentials
-        if: steps.gate.outputs.provider_mode == 'openai_direct' || steps.gate.outputs.provider_mode == 'openrouter'
+        if: steps.gate.outputs.provider_mode == 'openai_direct' || steps.gate.outputs.provider_mode == 'openrouter' || steps.gate.outputs.provider_mode == 'nvidia_nim'
         env:
           GITHUB_MODELS_FALLBACK_TOKEN: ${{ secrets.STRIX_GITHUB_MODELS_TOKEN || github.token }}
         run: |
@@ -680,7 +729,7 @@ jobs:
       - name: Prepare Strix model input file
         if: steps.gate.outputs.enabled == 'true'
         env:
-          STRIX_MODEL: ${{ github.event.client_payload.strix_llm || 'gpt-5.6-luna' }}
+          STRIX_MODEL: ${{ github.event.client_payload.strix_llm || (steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna') }}
         run: |
           umask 077
           strix_llm_file="$RUNNER_TEMP/strix_llm.txt"
@@ -709,11 +758,14 @@ jobs:
             openrouter/free | openrouter/openrouter/free)
               printf '%s' 'openrouter/free' > "$strix_llm_file"
               ;;
+            nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b)
+              printf '%s' "$strix_model" > "$strix_llm_file"
+              ;;
             vertex_ai/gemini-3.1-pro-preview-customtools | vertex_ai/gemini-2.5-flash)
               printf '%s' "$strix_model" > "$strix_llm_file"
               ;;
             *)
-              echo '::error::STRIX_LLM must select GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model.'
+              echo '::error::STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model.'
               exit 1
               ;;
           esac
@@ -732,7 +784,7 @@ jobs:
           STRIX_LLM_FILE: ${{ env.STRIX_LLM_FILE }}
           STRIX_REPO_ROOT: ${{ runner.temp }}/trusted-workspace
           LLM_API_BASE_FILE: ${{ env.LLM_API_BASE_FILE }}
-          STRIX_LLM_DEFAULT_PROVIDER: ${{ steps.gate.outputs.provider_mode == 'vertex_ai' && 'vertex_ai' || 'openai' }}
+          STRIX_LLM_DEFAULT_PROVIDER: ${{ steps.gate.outputs.provider_mode == 'vertex_ai' && 'vertex_ai' || steps.gate.outputs.provider_mode == 'nvidia_nim' && 'nvidia_nim' || 'openai' }}
           LLM_API_KEY_FILE: ${{ env.LLM_API_KEY_FILE }}
           GOOGLE_APPLICATION_CREDENTIALS: ${{ env.GOOGLE_APPLICATION_CREDENTIALS }}
           CLOUDSDK_AUTH_CREDENTIAL_FILE_OVERRIDE: ${{ env.CLOUDSDK_AUTH_CREDENTIAL_FILE_OVERRIDE }}
@@ -750,7 +802,7 @@ jobs:
           STRIX_LLM_MAX_RETRIES: 1
           STRIX_TRANSIENT_RETRY_PER_MODEL: 2
           STRIX_TRANSIENT_RETRY_BACKOFF_SECONDS: 60
-          STRIX_FALLBACK_MODELS: ${{ steps.gate.outputs.provider_mode == 'github_models' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || steps.gate.outputs.provider_mode == 'openai_direct' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || steps.gate.outputs.provider_mode == 'openrouter' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || '' }}
+          STRIX_FALLBACK_MODELS: ${{ steps.gate.outputs.provider_mode == 'github_models' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || steps.gate.outputs.provider_mode == 'openai_direct' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || steps.gate.outputs.provider_mode == 'openrouter' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || steps.gate.outputs.provider_mode == 'nvidia_nim' && 'github_models/openai/o3 github_models/openai/gpt-5-chat' || '' }}
           STRIX_GITHUB_MODELS_API_BASE_FILE: ${{ env.STRIX_GITHUB_MODELS_API_BASE_FILE }}
           STRIX_GITHUB_MODELS_KEY_FILE: ${{ env.STRIX_GITHUB_MODELS_KEY_FILE }}
           STRIX_FAIL_ON_PROVIDER_SIGNAL: "1"
diff --git a/scripts/ci/emit_opencode_failed_check_fallback_findings.sh b/scripts/ci/emit_opencode_failed_check_fallback_findings.sh
index c6db1a2db..ccd35a273 100755
--- a/scripts/ci/emit_opencode_failed_check_fallback_findings.sh
+++ b/scripts/ci/emit_opencode_failed_check_fallback_findings.sh
@@ -956,13 +956,13 @@ extract_strix_failed_check_block "$EVIDENCE_FILE" "$strix_evidence_file"
 
 emit_known_missing_string_finding \
 	"$EVIDENCE_FILE" \
-	"github.event.client_payload.strix_llm || 'gpt-5.6-luna'" \
-	"Strix PR scans must default to direct OpenAI GPT-5.6 Luna" \
+	"steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna'" \
+	"Strix public scans must default to NVIDIA NIM while private scans retain the contracted provider" \
 	".github/workflows/strix.yml" \
 	"scripts/ci/test_strix_quick_gate.sh"
 emit_known_missing_string_finding \
 	"$EVIDENCE_FILE" \
-	"STRIX_LLM must select GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" \
+	"STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" \
 	"Strix unsupported-model errors must name the allowed providers" \
 	".github/workflows/strix.yml" \
 	"scripts/ci/test_strix_quick_gate.sh"
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 389900d53..08e31a6a7 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -226,6 +226,8 @@ assert_strix_workflow_pr_trigger_hardened() {
 	assert_file_contains "$workflow_file" "Fetch pull request head for trusted scan" "strix workflow fetches PR head without checkout"
 	assert_file_contains "$workflow_file" "github.event.client_payload.pr_number" "strix workflow consumes default-branch PR-scope evidence payloads"
 	assert_file_contains "$workflow_file" "github.event.client_payload.strix_llm" "strix workflow accepts only repository-dispatch Strix model overrides"
+	assert_file_contains "$workflow_file" "Resolve target repository visibility" "strix workflow resolves target privacy before selecting hosted trial providers"
+	assert_file_contains "$workflow_file" "NVIDIA NIM hosted trial scans are limited to public repositories" "strix workflow blocks NVIDIA hosted trial scans for private repositories"
 	assert_file_contains "$workflow_file" "github.event.client_payload.pr_number" "strix workflow can run PR-scoped repository_dispatch evidence"
 	assert_file_contains "$workflow_file" "PR number and head SHA are required for trusted PR-scope Strix evidence" "strix workflow fails closed when manual PR-scope metadata is incomplete"
 	assert_file_contains "$workflow_file" '[[ "$PR_HEAD_SHA" =~ ^[0-9a-fA-F]{40}$ ]]' "strix workflow validates PR head SHA before trusted fetch"
@@ -278,9 +280,9 @@ assert_strix_workflow_pr_trigger_hardened() {
 	assert_file_not_contains "$workflow_file" "STRIX_TOTAL_TIMEOUT_SECONDS:" "strix workflow must not expose total timeout env names in GitHub logs"
 	assert_file_not_contains "$workflow_file" "STRIX_PR_SCOPE_MAX_FILES_PER_BATCH" "strix workflow must not split Strix PR evidence into separate scanner runs"
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM == 'vertex_ai/gemini-3.1-pro-preview-customtools' && 'vertex_ai/gemini-2.5-flash'" "strix workflow must not quarantine the approved Vertex preview model after organization secret visibility is fixed"
-	assert_file_contains "$workflow_file" "github.event.client_payload.strix_llm || 'gpt-5.6-luna'" "strix workflow defaults PR Strix scans to direct OpenAI GPT-5.6 Luna"
+	assert_file_contains "$workflow_file" "steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna'" "strix workflow defaults public scans to NVIDIA NIM and keeps private scans on the contracted provider"
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM ||" "strix workflow must not let the legacy STRIX_LLM secret override PR defaults"
-	assert_file_contains "$workflow_file" "STRIX_LLM must select GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" "strix workflow rejects unsupported model inputs"
+	assert_file_contains "$workflow_file" "STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" "strix workflow rejects unsupported model inputs"
 	assert_file_contains "$workflow_file" "vertex_ai/gemini-3.1-pro-preview-customtools | vertex_ai/gemini-2.5-flash)" "strix workflow accepts only exact approved organization Vertex AI models"
 	assert_file_contains "$workflow_file" 'STRIX_VERTEX_FALLBACK_MODELS: ""' "strix workflow disables silent Vertex fallbacks so timeout-class failures fail closed"
 	assert_file_contains "$workflow_file" 'STRIX_FAIL_ON_PROVIDER_SIGNAL: "1"' "strix workflow fails closed on timeout, fatal, warning, denied, or provider failure signals"
@@ -309,28 +311,33 @@ assert_strix_workflow_pr_trigger_hardened() {
 	assert_file_contains "$workflow_file" "provider_mode=openai_direct" "strix workflow requires direct OpenAI GPT-5 credentials"
 	assert_file_contains "$workflow_file" "provider_mode=github_models" "strix workflow supports GitHub Models provider mode"
 	assert_file_contains "$workflow_file" "provider_mode=openrouter" "strix workflow supports OpenRouter provider mode"
+	assert_file_contains "$workflow_file" "provider_mode=nvidia_nim" "strix workflow supports NVIDIA NIM provider mode"
 	assert_file_contains "$workflow_file" 'STRIX_GITHUB_MODELS_TOKEN: ${{ secrets.STRIX_GITHUB_MODELS_TOKEN || github.token }}' "strix workflow prefers the organization GitHub Models token secret and falls back to GITHUB_TOKEN"
 	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'github_models' && (secrets.STRIX_GITHUB_MODELS_TOKEN || github.token)" "strix workflow keeps GitHub Models key routing in provider-scoped key material"
 	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'openai_direct' && (secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY)" "strix workflow keeps direct OpenAI key routing in provider-scoped key material"
 	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'openrouter' && secrets.OPENROUTER_API_KEY" "strix workflow includes OpenRouter key routing in provider-scoped key material"
+	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'nvidia_nim' && secrets.NVIDIA_NIM_API_KEY" "strix workflow includes NVIDIA NIM key routing in provider-scoped key material"
 	assert_file_not_contains "$workflow_file" "secrets.LLM_API_KEY" "strix workflow must not expose generic LLM_API_KEY for Vertex scans"
 	assert_file_contains "$workflow_file" "STRIX_GITHUB_MODELS_TOKEN is required for GitHub Models Strix scans" "strix workflow fails closed when GitHub Models credentials are absent"
 	assert_file_contains "$workflow_file" "STRIX_OPENAI_API_KEY is required for Strix OpenAI Platform scans" "strix workflow fails closed when direct credentials are absent"
 	assert_file_contains "$workflow_file" "OPENROUTER_API_KEY is required for Strix OpenRouter scans" "strix workflow fails closed when OpenRouter credentials are absent"
+	assert_file_contains "$workflow_file" "NVIDIA_NIM_API_KEY is required for Strix NVIDIA NIM scans" "strix workflow fails closed when NVIDIA credentials are absent"
 	assert_file_contains "$workflow_file" 'PROVIDER_MODE: ${{ steps.gate.outputs.provider_mode }}' "strix workflow passes provider mode through env"
 	assert_file_not_contains "$workflow_file" '[ "${{ steps.gate.outputs.provider_mode }}" = "openai_direct" ]' "strix workflow does not interpolate provider mode inside shell condition"
 	assert_file_contains "$workflow_file" "STRIX_REASONING_EFFORT: high" "strix workflow uses high reasoning effort when the selected provider/model supports it"
 	assert_file_contains "$workflow_file" 'trimmed_openai_key="$(printf '"'"'%s'"'"' "$sanitized_openai_key" | sed '"'"'s/^[[:space:]]*//;s/[[:space:]]*$//'"'"')"' "strix workflow trims whitespace-only OpenAI keys before gate validation"
 	assert_file_contains "$workflow_file" 'printf '"'"'%s'"'"' "$trimmed" > "$llm_api_key_file"' "strix workflow writes trimmed provider API keys into the trusted input file"
-	assert_file_contains "$workflow_file" 'STRIX_LLM_DEFAULT_PROVIDER: ${{ steps.gate.outputs.provider_mode == '"'"'vertex_ai'"'"' && '"'"'vertex_ai'"'"' || '"'"'openai'"'"' }}' "strix workflow selects the correct default provider"
+	assert_file_contains "$workflow_file" 'STRIX_LLM_DEFAULT_PROVIDER: ${{ steps.gate.outputs.provider_mode == '"'"'vertex_ai'"'"' && '"'"'vertex_ai'"'"' || steps.gate.outputs.provider_mode == '"'"'nvidia_nim'"'"' && '"'"'nvidia_nim'"'"' || '"'"'openai'"'"' }}' "strix workflow selects the correct default provider"
 	assert_file_contains "$workflow_file" "Prepare GitHub Models API base" "strix workflow prepares the GitHub Models API base only for GitHub Models mode"
 	assert_file_contains "$workflow_file" "https://models.github.ai/inference" "strix workflow routes GitHub Models scans to the inference endpoint"
 	assert_file_contains "$workflow_file" "Prepare OpenRouter API base" "strix workflow prepares the OpenRouter API base when OpenRouter mode is selected"
 	assert_file_contains "$workflow_file" "https://openrouter.ai/api/v1" "strix workflow routes OpenRouter scans to the OpenRouter API endpoint"
+	assert_file_contains "$workflow_file" "https://integrate.api.nvidia.com/v1" "strix workflow routes NVIDIA NIM scans to the hosted endpoint"
 	assert_file_contains "$workflow_file" "LLM_API_BASE_FILE" "strix workflow passes the GitHub Models API base through a trusted input file"
 	assert_file_not_contains "$workflow_file" '${{ secrets.STRIX_OPENAI_API_KEY || github.token }}' "strix workflow must not use fallback-secret syntax for LLM API keys"
 	assert_file_contains "$workflow_file" "github_models/openai/o3 github_models/openai/gpt-5-chat" "strix workflow keeps GitHub Models fallback on tool-capable OpenAI models without GPT-4.1 downgrade"
 	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'openai_direct' && 'github_models/openai/o3 github_models/openai/gpt-5-chat'" "strix workflow gives direct-OpenAI scans GitHub Models fallbacks so provider quota outages degrade instead of skipping"
+	assert_file_contains "$workflow_file" "steps.gate.outputs.provider_mode == 'nvidia_nim' && 'github_models/openai/o3 github_models/openai/gpt-5-chat'" "strix workflow gives NVIDIA NIM scans contracted fallbacks"
 	assert_file_contains "$workflow_file" "Prepare GitHub Models fallback credentials" "strix workflow provisions GitHub Models fallback credentials for direct-OpenAI scans"
 	assert_file_contains "$GATE_SCRIPT" "STRIX_GITHUB_MODELS_KEY_FILE" "strix gate reads the optional GitHub Models fallback key file"
 	assert_file_contains "$GATE_SCRIPT" "STRIX_GITHUB_MODELS_API_BASE_FILE" "strix gate routes github_models fallback models through the GitHub Models endpoint"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index ccf946a74..05b641a9d 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -98,7 +98,8 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
         "opencode-free/laguna-s-2.1-free "
         "opencode-free/ling-3.0-flash-free "
         "opencode-free/big-pickle "
-        "opencode-free/mimo-v2.5-free ' || '' }}"
+        "opencode-free/mimo-v2.5-free "
+        "nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b ' || '' }}"
     )
     candidates_text = candidates_match.group(1)
     assert candidates_text.startswith(conditional_public_candidate)
@@ -110,6 +111,7 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
         "opencode-free/ling-3.0-flash-free",
         "opencode-free/big-pickle",
         "opencode-free/mimo-v2.5-free",
+        "nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
         *candidates_text.removeprefix(conditional_public_candidate).split(),
     ]
     candidate_pairs = [candidate.split("/", 1) for candidate in candidates]
@@ -126,6 +128,10 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
     ]
 
     assert candidate_pairs
+    assert all(
+        not candidate.startswith("nvidia-nim/")
+        for candidate in candidates_text.removeprefix(conditional_public_candidate).split()
+    )
     assert candidate_pairs == [
         ["opencode-free", "nemotron-3-ultra-free"],
         ["opencode-free", "deepseek-v4-flash-free"],
@@ -1314,7 +1320,8 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
         "opencode-free/laguna-s-2.1-free "
         "opencode-free/ling-3.0-flash-free "
         "opencode-free/big-pickle "
-        "opencode-free/mimo-v2.5-free ' || ''"
+        "opencode-free/mimo-v2.5-free "
+        "nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b ' || ''"
     ) in workflow
     assert (
         "github-models/deepseek/deepseek-v3-0324 "
diff --git a/tests/test_required_workflow_queue_contract.py b/tests/test_required_workflow_queue_contract.py
index 1ea6376fe..dae188cbd 100644
--- a/tests/test_required_workflow_queue_contract.py
+++ b/tests/test_required_workflow_queue_contract.py
@@ -376,6 +376,11 @@ def test_noema_review_credentials_and_llm_configuration_fail_closed() -> None:
         "NOEMA_LLM_API_KEY: ${{ secrets.NOEMA_LLM_API_KEY || secrets.OPENAI_API_KEY || '' }}"
         in workflow
     )
+    assert "Resolve Noema target repository visibility" in workflow
+    assert 'if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ]' in workflow
+    assert "https://integrate.api.nvidia.com/v1/chat/completions" in workflow
+    assert 'export NOEMA_LLM_MODEL="nvidia/nemotron-3-ultra-550b-a55b"' in workflow
+    assert "NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}" in workflow
     assert "Noema LLM is unconfigured:" in workflow
     assert "mark_unconfigured()" not in workflow
     assert "review skipped until Noema is deployed" not in workflow

From 8d7f9f3130767b60d9b7b091a69098e1d5ffede6 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Thu, 30 Jul 2026 18:35:51 +0900
Subject: [PATCH 03/10] fix(review): preserve provider fallbacks

---
 .github/workflows/noema-review.yml            |  2 +-
 .github/workflows/strix.yml                   |  4 ++
 scripts/ci/test_strix_quick_gate.sh           |  1 +
 .../test_required_workflow_queue_contract.py  | 71 ++++++++++++++++++-
 4 files changed, 76 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/noema-review.yml b/.github/workflows/noema-review.yml
index dd6156274..59b25e343 100644
--- a/.github/workflows/noema-review.yml
+++ b/.github/workflows/noema-review.yml
@@ -287,7 +287,7 @@ jobs:
             echo "::error::Noema reviewer credential selection succeeded but no token was minted; review cannot submit a verdict."
             exit 1
           fi
-          if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ] && [ -z "${NOEMA_LLM_API_URL:-}" ] && [ -z "${NOEMA_LLM_MODEL:-}" ]; then
+          if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ] && [ -n "${NVIDIA_NIM_API_KEY:-}" ] && [ -z "${NOEMA_LLM_API_URL:-}" ] && [ -z "${NOEMA_LLM_MODEL:-}" ]; then
             export NOEMA_LLM_API_URL="https://integrate.api.nvidia.com/v1/chat/completions"
             export NOEMA_LLM_MODEL="nvidia/nemotron-3-ultra-550b-a55b"
             export NOEMA_LLM_API_KEY="${NVIDIA_NIM_API_KEY:-}"
diff --git a/.github/workflows/strix.yml b/.github/workflows/strix.yml
index 4f95a31f6..da69668db 100644
--- a/.github/workflows/strix.yml
+++ b/.github/workflows/strix.yml
@@ -444,6 +444,7 @@ jobs:
         id: gate
         env:
           STRIX_MODEL: ${{ github.event.client_payload.strix_llm || (steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna') }}
+          STRIX_MODEL_REQUESTED: ${{ github.event.client_payload.strix_llm || '' }}
           STRIX_OPENAI_API_KEY: ${{ secrets.STRIX_OPENAI_API_KEY || secrets.OPENAI_API_KEY }}
           STRIX_OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
           STRIX_NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
@@ -452,6 +453,9 @@ jobs:
           TARGET_REPOSITORY_PRIVATE: ${{ steps.target_visibility.outputs.is_private }}
         run: |
           strix_model="$(printf '%s' "$STRIX_MODEL" | sed 's/^[[:space:]]*//;s/[[:space:]]*$//')"
+          if [ -z "$STRIX_MODEL_REQUESTED" ] && [ "$strix_model" = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b" ] && [ -z "${STRIX_NVIDIA_NIM_API_KEY:-}" ]; then
+            strix_model="gpt-5.6-luna"
+          fi
           case "$strix_model" in
             openai/gpt-5-mini* | openai/gpt-5-nano* | \
             openai/openai/gpt-5-mini* | openai/openai/gpt-5-nano* | \
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 08e31a6a7..594cb52a6 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -281,6 +281,7 @@ assert_strix_workflow_pr_trigger_hardened() {
 	assert_file_not_contains "$workflow_file" "STRIX_PR_SCOPE_MAX_FILES_PER_BATCH" "strix workflow must not split Strix PR evidence into separate scanner runs"
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM == 'vertex_ai/gemini-3.1-pro-preview-customtools' && 'vertex_ai/gemini-2.5-flash'" "strix workflow must not quarantine the approved Vertex preview model after organization secret visibility is fixed"
 	assert_file_contains "$workflow_file" "steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna'" "strix workflow defaults public scans to NVIDIA NIM and keeps private scans on the contracted provider"
+	assert_file_contains "$workflow_file" 'if [ -z "$STRIX_MODEL_REQUESTED" ] && [ "$strix_model" = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b" ] && [ -z "${STRIX_NVIDIA_NIM_API_KEY:-}" ]' "strix workflow falls back to the contracted provider when the NVIDIA secret is absent"
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM ||" "strix workflow must not let the legacy STRIX_LLM secret override PR defaults"
 	assert_file_contains "$workflow_file" "STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" "strix workflow rejects unsupported model inputs"
 	assert_file_contains "$workflow_file" "vertex_ai/gemini-3.1-pro-preview-customtools | vertex_ai/gemini-2.5-flash)" "strix workflow accepts only exact approved organization Vertex AI models"
diff --git a/tests/test_required_workflow_queue_contract.py b/tests/test_required_workflow_queue_contract.py
index dae188cbd..ce1962bd9 100644
--- a/tests/test_required_workflow_queue_contract.py
+++ b/tests/test_required_workflow_queue_contract.py
@@ -1,4 +1,5 @@
 import json
+import os
 import shlex
 import shutil
 import subprocess
@@ -377,7 +378,10 @@ def test_noema_review_credentials_and_llm_configuration_fail_closed() -> None:
         in workflow
     )
     assert "Resolve Noema target repository visibility" in workflow
-    assert 'if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ]' in workflow
+    assert (
+        'if [ "$TARGET_REPOSITORY_PRIVATE" = "false" ] && '
+        '[ -n "${NVIDIA_NIM_API_KEY:-}" ]'
+    ) in workflow
     assert "https://integrate.api.nvidia.com/v1/chat/completions" in workflow
     assert 'export NOEMA_LLM_MODEL="nvidia/nemotron-3-ultra-550b-a55b"' in workflow
     assert "NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}" in workflow
@@ -387,6 +391,71 @@ def test_noema_review_credentials_and_llm_configuration_fail_closed() -> None:
     assert "Noema app token is unavailable; review skipped." not in workflow
 
 
+def test_nvidia_nim_defaults_preserve_existing_fallbacks_without_secret(
+    tmp_path: Path,
+) -> None:
+    strix_output = tmp_path / "strix-output"
+    strix = subprocess.run(
+        [
+            "bash",
+            "-c",
+            textwrap.dedent(
+                workflow_step(workflow_text("strix.yml"), "Gate Strix secrets")
+                .split("        run: |\n", 1)[1]
+            ),
+        ],
+        env={
+            **os.environ,
+            "GITHUB_OUTPUT": str(strix_output),
+            "STRIX_MODEL": "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b",
+            "STRIX_MODEL_REQUESTED": "",
+            "STRIX_OPENAI_API_KEY": "synthetic-openai-key",
+            "STRIX_OPENROUTER_API_KEY": "",
+            "STRIX_NVIDIA_NIM_API_KEY": "",
+            "STRIX_VERTEX_CREDENTIALS": "",
+            "STRIX_GITHUB_MODELS_TOKEN": "synthetic-models-token",
+            "TARGET_REPOSITORY_PRIVATE": "false",
+        },
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert strix.returncode == 0, strix.stderr
+    assert "provider_mode=openai_direct" in strix_output.read_text()
+
+    noema_probe = tmp_path / "noema-key"
+    noema_script = textwrap.dedent(
+        workflow_step(
+            workflow_text("noema-review.yml"),
+            "Run Noema LLM review and submit verdict",
+        ).split("        run: |\n", 1)[1]
+    )
+    noema = subprocess.run(
+        [
+            "bash",
+            "-c",
+            f"trap 'printf %s \"$NOEMA_LLM_API_KEY\" > {shlex.quote(str(noema_probe))}' EXIT\n"
+            + noema_script,
+        ],
+        env={
+            **os.environ,
+            "PR_NUMBER": "1",
+            "GH_TOKEN": "synthetic-review-token",
+            "NOEMA_LLM_API_URL": "",
+            "NOEMA_LLM_MODEL": "",
+            "NOEMA_LLM_API_KEY": "synthetic-openai-key",
+            "NVIDIA_NIM_API_KEY": "",
+            "TARGET_REPOSITORY_PRIVATE": "false",
+        },
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert noema.returncode == 1
+    assert "Noema LLM is unconfigured" in noema.stdout
+    assert noema_probe.read_text() == "synthetic-openai-key"
+
+
 def test_noema_workflow_run_without_pull_request_skips_before_token_exchange() -> None:
     workflow = workflow_text("noema-review.yml")
 

From 45df7d28458f9afb8dcd3720905a264776c795e2 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Thu, 30 Jul 2026 18:43:36 +0900
Subject: [PATCH 04/10] fix(review): propagate Strix fallback model

---
 .github/workflows/strix.yml                    | 3 ++-
 scripts/ci/test_strix_quick_gate.sh            | 1 +
 tests/test_required_workflow_queue_contract.py | 9 ++++++++-
 3 files changed, 11 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/strix.yml b/.github/workflows/strix.yml
index da69668db..5c10afa51 100644
--- a/.github/workflows/strix.yml
+++ b/.github/workflows/strix.yml
@@ -456,6 +456,7 @@ jobs:
           if [ -z "$STRIX_MODEL_REQUESTED" ] && [ "$strix_model" = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b" ] && [ -z "${STRIX_NVIDIA_NIM_API_KEY:-}" ]; then
             strix_model="gpt-5.6-luna"
           fi
+          echo "strix_model=$strix_model" >> "$GITHUB_OUTPUT"
           case "$strix_model" in
             openai/gpt-5-mini* | openai/gpt-5-nano* | \
             openai/openai/gpt-5-mini* | openai/openai/gpt-5-nano* | \
@@ -733,7 +734,7 @@ jobs:
       - name: Prepare Strix model input file
         if: steps.gate.outputs.enabled == 'true'
         env:
-          STRIX_MODEL: ${{ github.event.client_payload.strix_llm || (steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna') }}
+          STRIX_MODEL: ${{ steps.gate.outputs.strix_model }}
         run: |
           umask 077
           strix_llm_file="$RUNNER_TEMP/strix_llm.txt"
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 594cb52a6..f7ce9dcfb 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -282,6 +282,7 @@ assert_strix_workflow_pr_trigger_hardened() {
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM == 'vertex_ai/gemini-3.1-pro-preview-customtools' && 'vertex_ai/gemini-2.5-flash'" "strix workflow must not quarantine the approved Vertex preview model after organization secret visibility is fixed"
 	assert_file_contains "$workflow_file" "steps.target_visibility.outputs.is_private == 'false' && 'nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b' || 'gpt-5.6-luna'" "strix workflow defaults public scans to NVIDIA NIM and keeps private scans on the contracted provider"
 	assert_file_contains "$workflow_file" 'if [ -z "$STRIX_MODEL_REQUESTED" ] && [ "$strix_model" = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b" ] && [ -z "${STRIX_NVIDIA_NIM_API_KEY:-}" ]' "strix workflow falls back to the contracted provider when the NVIDIA secret is absent"
+	assert_file_contains "$workflow_file" 'STRIX_MODEL: ${{ steps.gate.outputs.strix_model }}' "strix workflow propagates the gate-selected fallback model to the scanner"
 	assert_file_not_contains "$workflow_file" "secrets.STRIX_LLM ||" "strix workflow must not let the legacy STRIX_LLM secret override PR defaults"
 	assert_file_contains "$workflow_file" "STRIX_LLM must select NVIDIA NIM Nemotron, GitHub Models openai/gpt-5 or newer, direct OpenAI GPT-5.4 or newer, OpenRouter openrouter/free, or an approved organization Vertex AI model" "strix workflow rejects unsupported model inputs"
 	assert_file_contains "$workflow_file" "vertex_ai/gemini-3.1-pro-preview-customtools | vertex_ai/gemini-2.5-flash)" "strix workflow accepts only exact approved organization Vertex AI models"
diff --git a/tests/test_required_workflow_queue_contract.py b/tests/test_required_workflow_queue_contract.py
index ce1962bd9..15b3bafb4 100644
--- a/tests/test_required_workflow_queue_contract.py
+++ b/tests/test_required_workflow_queue_contract.py
@@ -421,7 +421,14 @@ def test_nvidia_nim_defaults_preserve_existing_fallbacks_without_secret(
         check=False,
     )
     assert strix.returncode == 0, strix.stderr
-    assert "provider_mode=openai_direct" in strix_output.read_text()
+    assert {
+        "provider_mode=openai_direct",
+        "strix_model=gpt-5.6-luna",
+    } <= set(strix_output.read_text().splitlines())
+    assert (
+        "STRIX_MODEL: ${{ steps.gate.outputs.strix_model }}"
+        in workflow_text("strix.yml")
+    )
 
     noema_probe = tmp_path / "noema-key"
     noema_script = textwrap.dedent(

From c9649649f9d8b3239cc7575bec52f9223202b865 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Fri, 31 Jul 2026 18:12:38 +0900
Subject: [PATCH 05/10] test: enforce Strix timeout invariants

---
 scripts/ci/test_strix_quick_gate.sh | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 1ceee93c1..9be340cbf 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -14,8 +14,15 @@ REPO_ROOT="$(
 GATE_SCRIPT="$REPO_ROOT/scripts/ci/strix_quick_gate.sh"
 
 FAILURES=0
-TIMEOUT_TEST_PROCESS_SECONDS="${STRIX_TEST_PROCESS_TIMEOUT_SECONDS:-10}"
-TIMEOUT_TEST_FAKE_SLEEP_SECONDS="${STRIX_TEST_FAKE_SLEEP_SECONDS:-30}"
+TIMEOUT_TEST_PROCESS_SECONDS="${STRIX_TEST_PROCESS_TIMEOUT_SECONDS:-30}"
+TIMEOUT_TEST_FAKE_SLEEP_SECONDS="${STRIX_TEST_FAKE_SLEEP_SECONDS:-60}"
+
+if ! [[ "$TIMEOUT_TEST_PROCESS_SECONDS" =~ ^[1-9][0-9]*$ ]] ||
+	! [[ "$TIMEOUT_TEST_FAKE_SLEEP_SECONDS" =~ ^[1-9][0-9]*$ ]] ||
+	[ "$TIMEOUT_TEST_FAKE_SLEEP_SECONDS" -le "$TIMEOUT_TEST_PROCESS_SECONDS" ]; then
+	printf 'STRIX_TEST_FAKE_SLEEP_SECONDS must be a positive integer greater than STRIX_TEST_PROCESS_TIMEOUT_SECONDS.\n' >&2
+	exit 2
+fi
 
 # Keep local developer/provider secrets from changing fake Strix model routing.
 unset STRIX_LLM
@@ -7898,7 +7905,7 @@ run_timeout_cleanup_case() {
 #!/usr/bin/env bash
 set -euo pipefail
 
-sleep 30 &
+sleep "${FAKE_STRIX_TIMEOUT_SLEEP_SECONDS:?}" &
 child_pid=$!
 printf '%s' "$child_pid" > "${FAKE_STRIX_CHILD_PID_FILE:?}"
 sleep "${FAKE_STRIX_TIMEOUT_SLEEP_SECONDS:?}"

From 5dc981ba9352beaee530b665f7eb5b4cfdf611a7 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Fri, 31 Jul 2026 19:18:59 +0900
Subject: [PATCH 06/10] fix: break OpenCode scheduler check cycle

---
 .github/workflows/opencode-review-dispatch.yml |  8 ++++++--
 scripts/ci/collect_failed_check_evidence.sh    |  2 +-
 scripts/ci/test_strix_quick_gate.sh            | 15 ++++++++++++---
 tests/test_opencode_agent_contract.py          |  3 ++-
 4 files changed, 21 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index 395176423..625aa7dc4 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -2459,6 +2459,7 @@ jobs:
                       | select((.name // "") != "Required OpenCode Review")
                       | select((.name // "") != "OpenCode PR Review")
                       | select((.name // "") != "metadata-only gate evaluation")
+                      | select((.name // "") != "scan-pr-queue")
                       | select((.checkSuite.workflowRun.workflow.name // "") != "OpenCode Review")
                       | select((.checkSuite.workflowRun.workflow.name // "") != "Required OpenCode Review")
                       | select((.checkSuite.workflowRun.workflow.name // "") != "OpenCode PR Review")
@@ -4376,7 +4377,7 @@ jobs:
           self_check_filter='
             def self_check:
               (.name // "") as $n
-              | ["opencode-review", "coverage-evidence", "coverage-source-tree", "required-workflow-bootstrap", "metadata-only gate evaluation"] | index($n);
+              | ["opencode-review", "coverage-evidence", "coverage-source-tree", "required-workflow-bootstrap", "metadata-only gate evaluation", "scan-pr-queue"] | index($n);
             def latest_peer_checks:
               [
                 (.check_runs // [])[]
@@ -6186,6 +6187,7 @@ jobs:
                   | select(((.name // "" | ascii_downcase) as $n | ["opencode-review","coverage-evidence","metadata-only gate evaluation"] | index($n)) | not)
                   | select((.status // "") == "completed")
                   | select((.conclusion // "" | ascii_upcase) as $c | ["FAILURE","TIMED_OUT","ACTION_REQUIRED","CANCELLED","STARTUP_FAILURE"] | index($c))
+                  | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
                   | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("$" + "{{"))) | not)
                   | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review") | not)
                   | "- " + (if (.name // "") == "strix" then "Strix Security Scan/strix" else ((.name // "check") + " check run") end) + ": " + (.conclusion // "unknown") + (if (.details_url // .html_url // "") != "" then " (" + (.details_url // .html_url) + ")" else "" end)
@@ -6199,6 +6201,7 @@ jobs:
                   | map(last)
                   | .[]?
                   | select(((.name // "" | ascii_downcase) as $n | ["opencode-review","coverage-evidence","metadata-only gate evaluation"] | index($n)) | not)
+                  | select((.name // "") != "scan-pr-queue")
                   | select((.status // "") != "completed")
                   | "- " + (if (.name // "") == "strix" then "Strix Security Scan/strix" else ((.name // "check") + " check run") end) + ": " + (.status // "unknown") + (if (.details_url // .html_url // "") != "" then " (" + (.details_url // .html_url) + ")" else "" end)
                 '
@@ -6498,7 +6501,7 @@ jobs:
                       | select((.conclusion // "" | ascii_upcase) as $c | ["FAILURE","TIMED_OUT","ACTION_REQUIRED","CANCELLED","STARTUP_FAILURE"] | index($c))
                       | select((.name // "") != "metadata-only gate evaluation")
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.isRequired // false) | not) and (.workflow // "") == "CodeQL") | not)
-                      | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue" and ((.workflow // "") == "PR Review Merge Scheduler" or (.workflow // "") == "Required PR Review Merge Scheduler")) | not)
+                      | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("$" + "{{"))) | not)
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review" and ((.workflow // "") == "Noema Review" or (.workflow // "") == "Required Noema Review")) | not)
                       | "- " + (.label // "check") + ": " + (.conclusion // "unknown") + (if (.detailsUrl // "") != "" then " (" + .detailsUrl + ")" else "" end)
@@ -6630,6 +6633,7 @@ jobs:
                       | select((.workflow // "") != "Required OpenCode Review")
                       | select((.workflow // "") != "OpenCode PR Review")
                       | select((.name // "") != "metadata-only gate evaluation")
+                      | select((.name // "") != "scan-pr-queue")
                       | select((.status // "") != "COMPLETED")
                       | "- " + (.label // "check") + ": " + (.status // "unknown") + (if (.detailsUrl // "") != "" then " (" + .detailsUrl + ")" else "" end)
                     elif .kind == "status" then
diff --git a/scripts/ci/collect_failed_check_evidence.sh b/scripts/ci/collect_failed_check_evidence.sh
index cc2e4033e..71c31d7d1 100755
--- a/scripts/ci/collect_failed_check_evidence.sh
+++ b/scripts/ci/collect_failed_check_evidence.sh
@@ -519,7 +519,7 @@ gh api graphql \
 				# so its stable check name is the only safe cycle-breaking key.
 				| select((.name // "") != "metadata-only gate evaluation")
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.isRequired // false) | not) and (.checkSuite.workflowRun.workflow.name // "") == "CodeQL") | not)
-				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue" and ((.checkSuite.workflowRun.workflow.name // "") == "PR Review Merge Scheduler" or (.checkSuite.workflowRun.workflow.name // "") == "Required PR Review Merge Scheduler")) | not)
+				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("${{"))) | not)
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review" and ((.checkSuite.workflowRun.workflow.name // "") == "Noema Review" or (.checkSuite.workflowRun.workflow.name // "") == "Required Noema Review")) | not)
 				| select((.name // "") != "opencode-review")
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 9be340cbf..c7fbbe2f0 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -1078,7 +1078,12 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$workflow_file" 'map(sort_by(.completedAt // "") | last)' "opencode approval considers only the latest completed statusCheckRollup entry per check label"
 	assert_file_contains "$workflow_file" '(.workflow // "") == "CodeQL"' "opencode approval can distinguish CodeQL dynamic setup checks"
 	assert_file_contains "$workflow_file" '((.isRequired // false) | not) and (.workflow // "") == "CodeQL"' "opencode approval ignores non-required cancelled CodeQL checks without source evidence"
-	assert_file_contains "$workflow_file" '(.name // "") == "scan-pr-queue" and ((.workflow // "") == "PR Review Merge Scheduler" or (.workflow // "") == "Required PR Review Merge Scheduler")' "opencode approval ignores cancelled scheduler queue replacement checks without source evidence"
+	assert_file_contains "$workflow_file" 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "opencode approval ignores cancelled scheduler queue replacement checks without workflow metadata"
+	scheduler_cancelled_filter_count="$(grep -Fc 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "$workflow_file")"
+	if [ "$scheduler_cancelled_filter_count" -lt 2 ]; then
+		fail "opencode GraphQL and commit-check fallback both ignore cancelled scheduler queue replacements (found ${scheduler_cancelled_filter_count}, expected at least 2)"
+	fi
+	assert_file_not_contains "$workflow_file" '(.name // "") == "scan-pr-queue" and ((.workflow // "") == "PR Review Merge Scheduler" or (.workflow // "") == "Required PR Review Merge Scheduler")' "opencode scheduler cancellation classification does not depend on optional workflow metadata"
 	assert_file_contains "$workflow_file" 'grep -Fq -- "Strix Security Scan/strix:" "$rollup_file"' "opencode approval avoids duplicate supplemental Strix workflow-run blockers when statusCheckRollup already has the Strix check"
 	assert_file_contains "$workflow_file" 'current_head_manual_strix_success_status()' "opencode approval can identify same-head manual Strix success status evidence"
 	assert_file_contains "$workflow_file" 'manual_run_line="$(latest_current_head_manual_strix_run || true)"' "opencode approval falls back to same-head manual Strix check-run success when commit status publication is unavailable"
@@ -1112,7 +1117,7 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'select((.name // "") != "metadata-only gate evaluation")' "failed-check evidence ignores metadata-only review-state gates even when GitHub misattributes their workflow"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'isRequired(pullRequestId: $prId)' "failed-check evidence reads PR-required status for check runs"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '((.isRequired // false) | not) and (.checkSuite.workflowRun.workflow.name // "") == "CodeQL"' "failed-check evidence ignores non-required cancelled CodeQL checks without logs"
-	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '(.name // "") == "scan-pr-queue" and ((.checkSuite.workflowRun.workflow.name // "") == "PR Review Merge Scheduler" or (.checkSuite.workflowRun.workflow.name // "") == "Required PR Review Merge Scheduler")' "failed-check evidence ignores cancelled scheduler queue replacement checks"
+	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "failed-check evidence ignores cancelled scheduler queue replacement checks without workflow metadata"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '((.name // "") | contains("${{"))' "failed-check evidence ignores cancelled matrix-template helper checks without logs"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '(.name // "") == "noema-review"' "failed-check evidence ignores cancelled Noema queue replacement checks without source logs"
 	assert_file_contains "$workflow_file" 'select((.name // "") != "metadata-only gate evaluation")' "opencode ignores metadata-only review-state gates without trusting GitHub workflow attribution"
@@ -1120,8 +1125,12 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	if [ "$metadata_gate_filter_count" -lt 3 ]; then
 		fail "opencode pre-model, failed-check, and pending-check collection all ignore metadata-only review-state gates (found ${metadata_gate_filter_count}, expected at least 3)"
 	fi
-	assert_file_contains "$workflow_file" '["opencode-review", "coverage-evidence", "coverage-source-tree", "required-workflow-bootstrap", "metadata-only gate evaluation"]' "central fast approval ignores its dependent metadata-only review-state gate"
+	assert_file_contains "$workflow_file" '["opencode-review", "coverage-evidence", "coverage-source-tree", "required-workflow-bootstrap", "metadata-only gate evaluation", "scan-pr-queue"]' "central fast approval ignores its dependent review and scheduler control-plane checks"
 	assert_file_contains "$workflow_file" '["opencode-review","coverage-evidence","metadata-only gate evaluation"]' "opencode supplemental check-run collection ignores review-state helper gates"
+	scheduler_pending_filter_count="$(grep -Fc 'select((.name // "") != "scan-pr-queue")' "$workflow_file")"
+	if [ "$scheduler_pending_filter_count" -lt 3 ]; then
+		fail "opencode pre-model, rollup, and commit-check pending collection all ignore the scheduler control-plane cycle (found ${scheduler_pending_filter_count}, expected at least 3)"
+	fi
 	assert_file_contains "$workflow_file" '((.name // "") | contains("$" + "{{"))' "opencode failed-check collection ignores cancelled matrix-template helper checks without logs without exposing a raw Actions expression"
 	assert_file_contains "$workflow_file" '(.name // "") == "noema-review"' "opencode failed-check collection ignores cancelled Noema queue replacement checks without source logs"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '"strix security scan/"*' "failed-check evidence maps stale Strix workflow helper checks to the manual strix evidence status"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index e1fb87c79..ea2a25646 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -1520,7 +1520,8 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     )
     assert (
         '["opencode-review", "coverage-evidence", "coverage-source-tree", '
-        '"required-workflow-bootstrap", "metadata-only gate evaluation"]' in workflow
+        '"required-workflow-bootstrap", "metadata-only gate evaluation", '
+        '"scan-pr-queue"]' in workflow
     )
     assert "falling back to current-head REST check-runs" in workflow
 

From 7030ed4fdf2b80d8aa884c0c4a8a52ac65f083c6 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Sat, 1 Aug 2026 04:41:09 +0900
Subject: [PATCH 07/10] fix(review): ignore scheduler self-check failures

---
 .../workflows/opencode-review-dispatch.yml    |  4 +--
 scripts/ci/collect_failed_check_evidence.sh   |  2 +-
 scripts/ci/test_strix_quick_gate.sh           | 10 +++---
 tests/test_opencode_agent_contract.py         | 34 +++++++++++++++++++
 4 files changed, 42 insertions(+), 8 deletions(-)

diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index dc10206c6..51a8cc0b2 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -6271,7 +6271,7 @@ jobs:
                   | select(((.name // "" | ascii_downcase) as $n | ["opencode-review","coverage-evidence","metadata-only gate evaluation"] | index($n)) | not)
                   | select((.status // "") == "completed")
                   | select((.conclusion // "" | ascii_upcase) as $c | ["FAILURE","TIMED_OUT","ACTION_REQUIRED","CANCELLED","STARTUP_FAILURE"] | index($c))
-                  | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
+                  | select((.name // "") != "scan-pr-queue")
                   | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("$" + "{{"))) | not)
                   | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review") | not)
                   | "- " + (if (.name // "") == "strix" then "Strix Security Scan/strix" else ((.name // "check") + " check run") end) + ": " + (.conclusion // "unknown") + (if (.details_url // .html_url // "") != "" then " (" + (.details_url // .html_url) + ")" else "" end)
@@ -6585,7 +6585,7 @@ jobs:
                       | select((.conclusion // "" | ascii_upcase) as $c | ["FAILURE","TIMED_OUT","ACTION_REQUIRED","CANCELLED","STARTUP_FAILURE"] | index($c))
                       | select((.name // "") != "metadata-only gate evaluation")
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.isRequired // false) | not) and (.workflow // "") == "CodeQL") | not)
-                      | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
+                      | select((.name // "") != "scan-pr-queue")
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("$" + "{{"))) | not)
                       | select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review" and ((.workflow // "") == "Noema Review" or (.workflow // "") == "Required Noema Review")) | not)
                       | "- " + (.label // "check") + ": " + (.conclusion // "unknown") + (if (.detailsUrl // "") != "" then " (" + .detailsUrl + ")" else "" end)
diff --git a/scripts/ci/collect_failed_check_evidence.sh b/scripts/ci/collect_failed_check_evidence.sh
index 71c31d7d1..51e5f1e5b 100755
--- a/scripts/ci/collect_failed_check_evidence.sh
+++ b/scripts/ci/collect_failed_check_evidence.sh
@@ -519,7 +519,7 @@ gh api graphql \
 				# so its stable check name is the only safe cycle-breaking key.
 				| select((.name // "") != "metadata-only gate evaluation")
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.isRequired // false) | not) and (.checkSuite.workflowRun.workflow.name // "") == "CodeQL") | not)
-				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)
+				| select((.name // "") != "scan-pr-queue")
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and ((.name // "") | contains("${{"))) | not)
 				| select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "noema-review" and ((.checkSuite.workflowRun.workflow.name // "") == "Noema Review" or (.checkSuite.workflowRun.workflow.name // "") == "Required Noema Review")) | not)
 				| select((.name // "") != "opencode-review")
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index f10505027..b1b306d39 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -1083,10 +1083,10 @@ assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" '
 	assert_file_contains "$workflow_file" 'map(sort_by(.completedAt // "") | last)' "opencode approval considers only the latest completed statusCheckRollup entry per check label"
 	assert_file_contains "$workflow_file" '(.workflow // "") == "CodeQL"' "opencode approval can distinguish CodeQL dynamic setup checks"
 	assert_file_contains "$workflow_file" '((.isRequired // false) | not) and (.workflow // "") == "CodeQL"' "opencode approval ignores non-required cancelled CodeQL checks without source evidence"
-	assert_file_contains "$workflow_file" 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "opencode approval ignores cancelled scheduler queue replacement checks without workflow metadata"
-	scheduler_cancelled_filter_count="$(grep -Fc 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "$workflow_file")"
-	if [ "$scheduler_cancelled_filter_count" -lt 2 ]; then
-		fail "opencode GraphQL and commit-check fallback both ignore cancelled scheduler queue replacements (found ${scheduler_cancelled_filter_count}, expected at least 2)"
+	assert_file_contains "$workflow_file" 'select((.name // "") != "scan-pr-queue")' "opencode approval ignores scheduler queue self-checks for every failed or pending state"
+	scheduler_self_check_filter_count="$(grep -Fc 'select((.name // "") != "scan-pr-queue")' "$workflow_file")"
+	if [ "$scheduler_self_check_filter_count" -lt 5 ]; then
+		fail "opencode GraphQL and commit-check failed/pending paths all ignore scheduler queue self-checks (found ${scheduler_self_check_filter_count}, expected at least 5)"
 	fi
 	assert_file_not_contains "$workflow_file" '(.name // "") == "scan-pr-queue" and ((.workflow // "") == "PR Review Merge Scheduler" or (.workflow // "") == "Required PR Review Merge Scheduler")' "opencode scheduler cancellation classification does not depend on optional workflow metadata"
 	assert_file_contains "$workflow_file" 'grep -Fq -- "Strix Security Scan/strix:" "$rollup_file"' "opencode approval avoids duplicate supplemental Strix workflow-run blockers when statusCheckRollup already has the Strix check"
@@ -1122,7 +1122,7 @@ assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" '
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'select((.name // "") != "metadata-only gate evaluation")' "failed-check evidence ignores metadata-only review-state gates even when GitHub misattributes their workflow"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'isRequired(pullRequestId: $prId)' "failed-check evidence reads PR-required status for check runs"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '((.isRequired // false) | not) and (.checkSuite.workflowRun.workflow.name // "") == "CodeQL"' "failed-check evidence ignores non-required cancelled CodeQL checks without logs"
-	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'select(((.conclusion // "" | ascii_downcase) == "cancelled" and (.name // "") == "scan-pr-queue") | not)' "failed-check evidence ignores cancelled scheduler queue replacement checks without workflow metadata"
+	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" 'select((.name // "") != "scan-pr-queue")' "failed-check evidence ignores scheduler queue self-checks for every failure conclusion"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '((.name // "") | contains("${{"))' "failed-check evidence ignores cancelled matrix-template helper checks without logs"
 	assert_file_contains "$REPO_ROOT/scripts/ci/collect_failed_check_evidence.sh" '(.name // "") == "noema-review"' "failed-check evidence ignores cancelled Noema queue replacement checks without source logs"
 	assert_file_contains "$workflow_file" 'select((.name // "") != "metadata-only gate evaluation")' "opencode ignores metadata-only review-state gates without trusting GitHub workflow attribution"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index 9d35523ef..e6d504beb 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -1619,6 +1619,40 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     assert "forced smooth scrolling" in prompt_template
 
 
+def test_opencode_excludes_queue_self_check_from_every_failed_check_path():
+    """Never diagnose the central scheduler's own queue check as a peer failure."""
+    workflow = Path(".github/workflows/opencode-review-dispatch.yml").read_text(
+        encoding="utf-8"
+    )
+    unconditional_filter = 'select((.name // "") != "scan-pr-queue")'
+    cancelled_only_filter = (
+        'select(((.conclusion // "" | ascii_downcase) == "cancelled" '
+        'and (.name // "") == "scan-pr-queue") | not)'
+    )
+
+    # Both failed-check collectors and both pending-check collectors exclude the
+    # scheduler check by name, independently of its current state or conclusion.
+    assert workflow.count(unconditional_filter) >= 5
+    assert cancelled_only_filter not in workflow
+    failed_check_collector = Path(
+        "scripts/ci/collect_failed_check_evidence.sh"
+    ).read_text(encoding="utf-8")
+    assert unconditional_filter in failed_check_collector
+    assert cancelled_only_filter not in failed_check_collector
+
+    fixtures = [
+        {"name": "scan-pr-queue", "conclusion": "CANCELLED"},
+        {"name": "scan-pr-queue", "conclusion": "FAILURE"},
+        {"name": "real-peer-check", "conclusion": "FAILURE"},
+    ]
+    retained = [
+        check
+        for check in fixtures
+        if check.get("name", "") != "scan-pr-queue"
+    ]
+    assert retained == [{"name": "real-peer-check", "conclusion": "FAILURE"}]
+
+
 def test_opencode_job_timeout_contains_full_sequential_review_budget():
     """Keep the outer job alive through evidence, review, and publication."""
     workflow = Path(".github/workflows/opencode-review-dispatch.yml").read_text(encoding="utf-8")

From 346a4be046091597b4bec0bdd194400aa1173a03 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Sat, 1 Aug 2026 05:03:31 +0900
Subject: [PATCH 08/10] fix(review): reserve fallback budget after NIM

---
 .../workflows/opencode-review-dispatch.yml    |  2 +
 scripts/ci/run_opencode_review_model_pool.sh  | 53 +++++++++++++++++--
 scripts/ci/test_strix_quick_gate.sh           |  4 ++
 tests/test_opencode_agent_contract.py         |  2 +
 tests/test_opencode_model_pool_runner.py      | 46 ++++++++++++++--
 5 files changed, 101 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index 51a8cc0b2..8c32616ed 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -4091,6 +4091,8 @@ jobs:
           OPENCODE_DYNAMIC_RUN_TIMEOUT_CAP_SECONDS: "5400"
           OPENCODE_DYNAMIC_TOTAL_BUDGET_CAP_SECONDS: "11700"
           OPENCODE_DYNAMIC_MAX_CYCLES_CAP: "1"
+          OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS: "180"
+          OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS: "900"
           OPENCODE_FREE_RUN_TIMEOUT_SECONDS: "3600"
           # This installation currently reports a 4k request-body limit for
           # GitHub Models GPT-5 endpoints even though the public catalog is
diff --git a/scripts/ci/run_opencode_review_model_pool.sh b/scripts/ci/run_opencode_review_model_pool.sh
index 6dcbd0a14..c92288d63 100644
--- a/scripts/ci/run_opencode_review_model_pool.sh
+++ b/scripts/ci/run_opencode_review_model_pool.sh
@@ -400,7 +400,10 @@ cap_model_run_timeout() {
 	local cap_seconds
 
 	case "$model_candidate" in
-	opencode-free/* | nvidia-nim/*)
+	nvidia-nim/*)
+		cap_seconds="$(env_integer_or_default OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS 180)"
+		;;
+	opencode-free/*)
 		cap_seconds="$(env_integer_or_default OPENCODE_FREE_RUN_TIMEOUT_SECONDS 3600)"
 		;;
 	github-models/openai/gpt-5 | github-models/openai/gpt-5-chat)
@@ -519,6 +522,8 @@ main() {
 	local uncapped_run_timeout
 	local changed_file_count small_file_threshold medium_file_threshold
 	local invalid_control_cap max_total_attempts total_attempts alive_candidates
+	local nim_budget_seconds nim_elapsed_seconds nim_remaining_seconds
+	local nim_attempt_started nim_attempt_elapsed non_nim_candidate_count
 	local -A dead_candidate_reasons invalid_control_counts
 	local -a model_candidates
 
@@ -582,8 +587,23 @@ main() {
 		fi
 		exit 1
 	fi
-	printf 'Configured OpenCode model pool: candidates=%s attempts=%s per-model-timeout=%ss retry-budget=%ss max-cycles=%s.\n' \
-		"${#model_candidates[@]}" "$attempts" "$original_run_timeout" "$budget_seconds" "$max_cycles"
+	nim_budget_seconds="$(env_integer_or_default OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS 900)"
+	nim_elapsed_seconds=0
+	non_nim_candidate_count=0
+	for model_candidate in "${model_candidates[@]}"; do
+		if ! is_nvidia_nim_candidate "$model_candidate"; then
+			non_nim_candidate_count=$((non_nim_candidate_count + 1))
+		fi
+	done
+	if [ "$non_nim_candidate_count" -gt 0 ] &&
+		[ "$budget_seconds" -gt 0 ] &&
+		[ "$nim_budget_seconds" -ge "$budget_seconds" ]; then
+		nim_budget_seconds=$((budget_seconds / 2))
+		printf 'OpenCode NVIDIA NIM combined runtime budget was capped at %ss so %s non-NIM fallback candidate(s) retain retry budget.\n' \
+			"$nim_budget_seconds" "$non_nim_candidate_count"
+	fi
+	printf 'Configured OpenCode model pool: candidates=%s attempts=%s per-model-timeout=%ss retry-budget=%ss max-cycles=%s NVIDIA-NIM-combined-budget=%ss.\n' \
+		"${#model_candidates[@]}" "$attempts" "$original_run_timeout" "$budget_seconds" "$max_cycles" "$nim_budget_seconds"
 
 	cycle=1
 	while :; do
@@ -597,6 +617,12 @@ main() {
 			if should_skip_model_candidate "$model_candidate"; then
 				continue
 			fi
+			if is_nvidia_nim_candidate "$model_candidate" &&
+				[ "$nim_elapsed_seconds" -ge "$nim_budget_seconds" ]; then
+				printf 'Skipping OpenCode %s because the NVIDIA NIM combined runtime budget of %ss is exhausted; preserving the remaining retry budget for fallback candidates.\n' \
+					"$model_candidate" "$nim_budget_seconds"
+				continue
+			fi
 			assert_reasoning_effort_for_candidate "$model_candidate"
 			safe_model="${model_candidate//[\/:]/-}"
 			prompt_file="${RUNNER_TEMP}/opencode-review-${safe_model}-prompt.md"
@@ -606,6 +632,12 @@ main() {
 			write_prompt "$model_candidate" "$prompt_file"
 			for attempt in $(seq 1 "$attempts"); do
 				now="$SECONDS"
+				if is_nvidia_nim_candidate "$model_candidate" &&
+					[ "$nim_elapsed_seconds" -ge "$nim_budget_seconds" ]; then
+					printf 'Stopping OpenCode %s retries because the NVIDIA NIM combined runtime budget of %ss is exhausted.\n' \
+						"$model_candidate" "$nim_budget_seconds"
+					break
+				fi
 				if [ "$deadline" -gt 0 ] && [ "$now" -ge "$deadline" ]; then
 					printf 'OpenCode model pool retry deadline elapsed before %s attempt %s/%s.\n' "$model_candidate" "$attempt" "$attempts"
 					if finish_pool_without_model; then
@@ -629,6 +661,14 @@ main() {
 				if [ "$deadline" -gt 0 ] && [ "$OPENCODE_RUN_TIMEOUT_SECONDS" -gt "$remaining" ]; then
 					OPENCODE_RUN_TIMEOUT_SECONDS="$remaining"
 				fi
+				if is_nvidia_nim_candidate "$model_candidate"; then
+					nim_remaining_seconds=$((nim_budget_seconds - nim_elapsed_seconds))
+					if [ "$OPENCODE_RUN_TIMEOUT_SECONDS" -gt "$nim_remaining_seconds" ]; then
+						printf 'OpenCode %s combined NVIDIA NIM budget cap selected %ss instead of %ss so fallback candidates retain retry budget.\n' \
+							"$model_candidate" "$nim_remaining_seconds" "$OPENCODE_RUN_TIMEOUT_SECONDS"
+						OPENCODE_RUN_TIMEOUT_SECONDS="$nim_remaining_seconds"
+					fi
+				fi
 				uncapped_run_timeout="$OPENCODE_RUN_TIMEOUT_SECONDS"
 				OPENCODE_RUN_TIMEOUT_SECONDS="$(cap_model_run_timeout "$model_candidate" "$OPENCODE_RUN_TIMEOUT_SECONDS")"
 				if [ "$OPENCODE_RUN_TIMEOUT_SECONDS" -lt "$uncapped_run_timeout" ]; then
@@ -642,6 +682,7 @@ main() {
 					agent="$OPENCODE_FIRST_ATTEMPT_AGENT"
 				fi
 				run_status=0
+				nim_attempt_started="$SECONDS"
 				if run_one_model_attempt "$model_candidate" "$attempt" "$attempts" "$agent" "$prompt_file" "$candidate_output_file" "$opencode_json_file" "$opencode_export_file"; then
 					cp "$candidate_output_file" "$OPENCODE_OUTPUT_FILE"
 					record_review_model "$model_candidate"
@@ -650,6 +691,12 @@ main() {
 				else
 					run_status=$?
 				fi
+				if is_nvidia_nim_candidate "$model_candidate"; then
+					nim_attempt_elapsed=$((SECONDS - nim_attempt_started))
+					nim_elapsed_seconds=$((nim_elapsed_seconds + nim_attempt_elapsed))
+					printf 'OpenCode NVIDIA NIM combined runtime used %ss/%ss after %s attempt %s/%s.\n' \
+						"$nim_elapsed_seconds" "$nim_budget_seconds" "$model_candidate" "$attempt" "$attempts"
+				fi
 				if [ "$run_status" -ne 3 ] && is_credit_exhausted_failure "$opencode_json_file" "${opencode_json_file}.stderr"; then
 					dead_candidate_reasons[$model_candidate]="provider credits exhausted (HTTP 402 / payment required)"
 					printf 'OpenCode %s provider credits are exhausted; marking this candidate failed for the rest of the run so retries cannot accrue further spend.\n' "$model_candidate"
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index b1b306d39..30adfcb59 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -731,9 +731,13 @@ assert_opencode_review_uses_codegraph_and_gpt5_fallback() {
 	assert_file_contains "$workflow_file" 'continue-on-error: true' "opencode approval gate still runs after model-pool failure to publish a reason"
 	assert_file_contains "$workflow_file" 'OPENCODE_RUN_TIMEOUT_SECONDS: "5400"' "opencode primary review preserves legitimate full-hour provider sessions"
 assert_file_contains "$workflow_file" 'OPENCODE_FREE_RUN_TIMEOUT_SECONDS: "3600"' "opencode free-tier failover timeout is hour-class (~3600s)"
+assert_file_contains "$workflow_file" 'OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS: "180"' "opencode NVIDIA NIM candidates have a short per-candidate failover timeout"
+assert_file_contains "$workflow_file" 'OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS: "900"' "opencode NVIDIA NIM candidates share a bounded combined runtime budget"
 assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_RUN_TIMEOUT_SECONDS:-3600' "opencode pool defaults primary run timeout to hour-class (~3600s) for large repos"
 assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_DYNAMIC_RUN_TIMEOUT_CAP_SECONDS 3600' "opencode pool dynamic timeout cap defaults to hour-class (~3600s)"
 assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_FREE_RUN_TIMEOUT_SECONDS 3600' "opencode free-tier failover timeout is hour-class (~3600s)"
+assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS 180' "opencode NVIDIA NIM candidate runtime cap defaults to three minutes"
+assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS 900' "opencode NVIDIA NIM combined runtime cap defaults to fifteen minutes"
 
 	assert_file_contains "$workflow_file" 'OPENCODE_TOTAL_RETRY_BUDGET_SECONDS: "11700"' "opencode model pool exits before the step timeout so the approval gate can publish a reason"
 	assert_file_contains "$workflow_file" 'OPENCODE_POOL_MAX_CYCLES: "1"' "opencode model pool exhausts each candidate only once before bounded fallback"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index e6d504beb..9a236d161 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -1415,6 +1415,8 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     assert 'OPENCODE_DYNAMIC_RUN_TIMEOUT_CAP_SECONDS: "5400"' in workflow
     assert 'OPENCODE_DYNAMIC_TOTAL_BUDGET_CAP_SECONDS: "11700"' in workflow
     assert 'OPENCODE_DYNAMIC_MAX_CYCLES_CAP: "1"' in workflow
+    assert 'OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS: "180"' in workflow
+    assert 'OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS: "900"' in workflow
     assert 'OPENCODE_FREE_RUN_TIMEOUT_SECONDS: "3600"' in workflow
     assert 'OPENCODE_GITHUB_GPT5_RUN_TIMEOUT_SECONDS: "45"' in workflow
     assert 'OPENCODE_DYNAMIC_MAX_CYCLES: "1"' in workflow
diff --git a/tests/test_opencode_model_pool_runner.py b/tests/test_opencode_model_pool_runner.py
index ac3203eec..fd09efcdf 100644
--- a/tests/test_opencode_model_pool_runner.py
+++ b/tests/test_opencode_model_pool_runner.py
@@ -27,6 +27,10 @@
     "OPENCODE_EVIDENCE_FILE",
     "OPENCODE_REQUIRE_ADVERSARIAL_VALIDATION",
 }
+INHERITED_PROVIDER_CREDENTIAL_ENV = {
+    "NVIDIA_API_KEY",
+    "NVIDIA_NIM_API_KEY",
+}
 
 
 def bash_command() -> str:
@@ -170,7 +174,7 @@ def run_failed_model(
     fake_opencode.chmod(0o755)
     github_output = tmp_path / "github-output.txt"
     env = os.environ.copy()
-    for name in CENTRAL_FALLBACK_ENV:
+    for name in CENTRAL_FALLBACK_ENV | INHERITED_PROVIDER_CREDENTIAL_ENV:
         env.pop(name, None)
     env.update(
         {
@@ -796,8 +800,12 @@ def test_free_provider_runtime_cap_preserves_queue_budget(tmp_path: Path) -> Non
     ) in result.stdout
 
 
-def test_nvidia_nim_candidate_requires_key(tmp_path: Path) -> None:
+def test_nvidia_nim_candidate_requires_key(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
     """NVIDIA NIM is skipped cleanly when its scoped credential is unavailable."""
+    monkeypatch.setenv("NVIDIA_NIM_API_KEY", "ambient-scoped-key")
+    monkeypatch.setenv("NVIDIA_API_KEY", "ambient-provider-key")
     result = run_failed_model(
         tmp_path,
         model_candidates="nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
@@ -814,7 +822,7 @@ def test_nvidia_nim_runtime_cap_preserves_queue_budget(tmp_path: Path) -> None:
         tmp_path,
         extra_env={
             "NVIDIA_NIM_API_KEY": "fake-nvidia-key",
-            "OPENCODE_FREE_RUN_TIMEOUT_SECONDS": "3",
+            "OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS": "3",
             "OPENCODE_RUN_TIMEOUT_SECONDS": "9",
         },
         model_candidates="nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
@@ -827,6 +835,38 @@ def test_nvidia_nim_runtime_cap_preserves_queue_budget(tmp_path: Path) -> None:
     ) in result.stdout
 
 
+def test_nvidia_nim_combined_budget_preserves_fallback_attempt(
+    tmp_path: Path,
+) -> None:
+    """Timed-out NIM candidates cannot consume the fallback provider budget."""
+    result = run_failed_model(
+        tmp_path,
+        extra_env={
+            "FAKE_OPENCODE_HANG_SECONDS": "2",
+            "NVIDIA_NIM_API_KEY": "fake-nvidia-key",
+            "OPENCODE_FREE_RUN_TIMEOUT_SECONDS": "1",
+            "OPENCODE_NVIDIA_NIM_RUN_TIMEOUT_SECONDS": "1",
+            "OPENCODE_NVIDIA_NIM_TOTAL_BUDGET_SECONDS": "1",
+            "OPENCODE_RUN_TIMEOUT_SECONDS": "5",
+            "OPENCODE_TOTAL_RETRY_BUDGET_SECONDS": "6",
+        },
+        model_candidates=(
+            "nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b "
+            "nvidia-nim/nvidia/nemotron-3-super-120b-a12b "
+            "opencode-free/nemotron-3-ultra-free"
+        ),
+    )
+
+    assert result.returncode == 1
+    assert "OpenCode NVIDIA NIM combined runtime used" in result.stdout
+    assert (
+        "Skipping OpenCode nvidia-nim/nvidia/nemotron-3-super-120b-a12b "
+        "because the NVIDIA NIM combined runtime budget of 1s is exhausted"
+        in result.stdout
+    )
+    assert "OpenCode opencode-free/nemotron-3-ultra-free attempt 1/1" in result.stdout
+
+
 def test_github_models_openai_prompt_references_evidence_without_inlining(
     tmp_path: Path,
 ) -> None:

From 8a8ac4123b242dd62aee4762f1a133f0f3b7144a Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Sat, 1 Aug 2026 05:45:28 +0900
Subject: [PATCH 09/10] fix(review): enforce scoped NIM credentials

---
 .../workflows/opencode-review-dispatch.yml    | 27 ++++++++++---------
 scripts/ci/run_opencode_review_model_pool.sh  | 12 +++++----
 scripts/ci/test_strix_quick_gate.sh           |  7 ++---
 tests/test_opencode_agent_contract.py         | 24 +++++++++++++----
 tests/test_opencode_model_pool_runner.py      |  3 ++-
 5 files changed, 46 insertions(+), 27 deletions(-)

diff --git a/.github/workflows/opencode-review-dispatch.yml b/.github/workflows/opencode-review-dispatch.yml
index 8c32616ed..885257e5f 100644
--- a/.github/workflows/opencode-review-dispatch.yml
+++ b/.github/workflows/opencode-review-dispatch.yml
@@ -4034,22 +4034,22 @@ jobs:
           # in the opencode.jsonc "openai" provider block.
           OPENCODE_API_KEY: ${{ secrets.OPENCODE_ZEN_API_KEY }}
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          # Org secret NVIDIA_NIM_API_KEY preferred; fallback NVIDIA_API_KEY.
-          # opencode.jsonc expects env NVIDIA_API_KEY for nvidia-nim/* models.
-          NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY }}
+          # The scoped NVIDIA_NIM_API_KEY is the only NIM credential source.
+          # opencode.jsonc expects that same scoped value in NVIDIA_API_KEY.
+          NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
           NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
           SHARE: "false"
           NPM_CONFIG_IGNORE_SCRIPTS: "true"
           NO_COLOR: "1"
           # High-sensitivity review candidates only. Public repositories first
-          # use OpenCode Zen's anonymous active, zero-cost model endpoints.
-          # Their trial/free-period data
-          # may be logged, retained, or used for product/model improvement, so
-          # private repositories never include them. The pool then falls
-          # through to
-          # NVIDIA NIM, OpenCode Zen GPT-5.6 Terra, DeepSeek V3, the direct
-          # GPT-5.6 Luna slot, and pinned PAID
+          # try NVIDIA NIM when its scoped secret is available, then OpenCode
+          # Zen's anonymous active, zero-cost models, followed by the existing
+          # provider fallbacks. Trial/free-period data may be logged, retained,
+          # or used for product/model improvement, so private repositories
+          # include neither NIM nor anonymous free candidates and start at the
+          # existing keyed fallback list: OpenCode Zen GPT-5.6 Terra, DeepSeek
+          # V3, the direct GPT-5.6 Luna slot, and pinned PAID
           # OpenRouter coder models (free-tier candidates hit the shared
           # free-models-per-day cap and hung for the full candidate timeout,
           # so the OpenRouter slots use cheap paid models billed against the
@@ -4695,9 +4695,10 @@ jobs:
           # Exposed so the "openai" provider in opencode.jsonc resolves during the
           # failed-check diagnosis opencode run that shares this config.
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          # Org secret NVIDIA_NIM_API_KEY preferred; fallback NVIDIA_API_KEY.
-          # opencode.jsonc expects env NVIDIA_API_KEY for nvidia-nim/* models.
-          NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY }}
+          # The scoped NVIDIA_NIM_API_KEY is the only NIM credential source.
+          # opencode.jsonc expects that same scoped value in NVIDIA_API_KEY.
+          NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
+          NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}
           OPENCODE_APP_TOKEN: ${{ steps.opencode_app_token.outputs.token }}
           OPENCODE_EVIDENCE_FILE: ${{ runner.temp }}/opencode-review-evidence.md
           OPENCODE_FAILED_CHECK_EVIDENCE_FILE: ${{ runner.temp }}/opencode-failed-check-evidence.md
diff --git a/scripts/ci/run_opencode_review_model_pool.sh b/scripts/ci/run_opencode_review_model_pool.sh
index c92288d63..8aef2b4d7 100644
--- a/scripts/ci/run_opencode_review_model_pool.sh
+++ b/scripts/ci/run_opencode_review_model_pool.sh
@@ -354,10 +354,12 @@ is_nvidia_nim_candidate() {
 
 # Org secret name is NVIDIA_NIM_API_KEY (GitHub Actions / org secrets UI).
 # opencode.jsonc nvidia-nim provider block resolves {env:NVIDIA_API_KEY}.
-# Workflow maps secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY → env NVIDIA_API_KEY.
-# Normalize here too so local/CLI runs with only NVIDIA_NIM_API_KEY set do not skip nim/*.
-if [ -z "${NVIDIA_API_KEY:-}" ] && [ -n "${NVIDIA_NIM_API_KEY:-}" ]; then
+# Normalize only the scoped secret and discard any legacy provider credential so
+# it cannot activate NIM candidates outside the explicit governance boundary.
+if [ -n "${NVIDIA_NIM_API_KEY:-}" ]; then
 	export NVIDIA_API_KEY="$NVIDIA_NIM_API_KEY"
+else
+	unset NVIDIA_API_KEY
 fi
 
 is_low_sensitivity_candidate() {
@@ -387,8 +389,8 @@ should_skip_model_candidate() {
 		printf 'Skipping OpenCode %s because OPENROUTER_API_KEY is not configured; falling back to the next provider-qualified candidate.\n' "$model_candidate"
 		return 0
 	fi
-	if is_nvidia_nim_candidate "$model_candidate" && [ -z "${NVIDIA_API_KEY:-}" ]; then
-		printf 'Skipping OpenCode %s because NVIDIA_API_KEY is not configured; falling back to the next provider-qualified candidate.\n' "$model_candidate"
+	if is_nvidia_nim_candidate "$model_candidate" && [ -z "${NVIDIA_NIM_API_KEY:-}" ]; then
+		printf 'Skipping OpenCode %s because scoped NVIDIA_NIM_API_KEY is not configured; falling back to the next provider-qualified candidate.\n' "$model_candidate"
 		return 0
 	fi
 	return 1
diff --git a/scripts/ci/test_strix_quick_gate.sh b/scripts/ci/test_strix_quick_gate.sh
index 30adfcb59..395c44b01 100755
--- a/scripts/ci/test_strix_quick_gate.sh
+++ b/scripts/ci/test_strix_quick_gate.sh
@@ -891,7 +891,7 @@ assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" '
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OpenCode model pool has no configured model candidates." "opencode model pool fails fast when no candidates are configured"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OPENAI_API_KEY is not configured" "opencode model pool skips native OpenAI candidates when the org secret is absent"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "OPENROUTER_API_KEY is not configured" "opencode model pool skips OpenRouter candidates when the org secret is absent"
-	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "NVIDIA_API_KEY is not configured" "opencode model pool skips NVIDIA NIM candidates when the normalized provider credential is absent"
+	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "scoped NVIDIA_NIM_API_KEY is not configured" "opencode model pool skips NVIDIA NIM candidates when the scoped credential is absent"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "configured max cycle count" "opencode model pool exits before the job timeout after configured cycles"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" 'OPENCODE_TOTAL_RETRY_BUDGET_SECONDS:-1500' "opencode model pool keeps a bounded default retry budget unless the workflow explicitly disables it"
 	assert_file_not_contains "$workflow_file" "no model produced a valid review control block" "opencode model-failure path no longer documents a final exhausted state"
@@ -1090,7 +1090,7 @@ assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" '
 	assert_file_contains "$workflow_file" 'select((.name // "") != "scan-pr-queue")' "opencode approval ignores scheduler queue self-checks for every failed or pending state"
 	scheduler_self_check_filter_count="$(grep -Fc 'select((.name // "") != "scan-pr-queue")' "$workflow_file")"
 	if [ "$scheduler_self_check_filter_count" -lt 5 ]; then
-		fail "opencode GraphQL and commit-check failed/pending paths all ignore scheduler queue self-checks (found ${scheduler_self_check_filter_count}, expected at least 5)"
+		record_failure "opencode GraphQL and commit-check failed/pending paths all ignore scheduler queue self-checks (found ${scheduler_self_check_filter_count}, expected at least 5)"
 	fi
 	assert_file_not_contains "$workflow_file" '(.name // "") == "scan-pr-queue" and ((.workflow // "") == "PR Review Merge Scheduler" or (.workflow // "") == "Required PR Review Merge Scheduler")' "opencode scheduler cancellation classification does not depend on optional workflow metadata"
 	assert_file_contains "$workflow_file" 'grep -Fq -- "Strix Security Scan/strix:" "$rollup_file"' "opencode approval avoids duplicate supplemental Strix workflow-run blockers when statusCheckRollup already has the Strix check"
@@ -1242,7 +1242,8 @@ assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" '
 	assert_file_contains "$workflow_file" "opencode/gpt-5.6-terra github-models/deepseek/deepseek-v3-0324 openai/gpt-5.6-luna openrouter/deepseek/deepseek-v3.2 openrouter/qwen/qwen3-coder github-models/openai/gpt-4.1 github-models/openai/gpt-5" "opencode review keeps paid Zen, DeepSeek V3, and full-size GPT fallbacks"
 	assert_file_contains "$workflow_file" "github-models/deepseek/deepseek-r1-0528" "opencode review keeps a reachable DeepSeek R1 reasoning fallback model"
 	assert_file_contains "$workflow_file" "github-models/deepseek/deepseek-v3-0324" "opencode review has a reachable DeepSeek V3 fallback model"
-	assert_file_contains "$workflow_file" "secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY" "opencode review binds org NVIDIA_NIM_API_KEY into NVIDIA_API_KEY env"
+	assert_file_not_contains "$workflow_file" "secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY" "opencode review never falls back from the scoped NVIDIA NIM secret to the legacy provider secret"
+	assert_file_contains "$workflow_file" 'NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}' "opencode review binds only the scoped NVIDIA NIM secret into the provider environment"
 	assert_file_contains "$REPO_ROOT/scripts/ci/run_opencode_review_model_pool.sh" "NVIDIA_NIM_API_KEY" "model pool normalizes NVIDIA_NIM_API_KEY to NVIDIA_API_KEY"
 
 	assert_file_contains "$workflow_file" "github-models/openai/gpt-5" "opencode review still has a bounded GPT-5 fallback model"
diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index 9a236d161..a1e45a1ab 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -198,6 +198,11 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
     assert nvidia_provider["models"]["nvidia/nemotron-3-ultra-550b-a55b"][
         "limit"
     ] == {"context": 131072, "output": 8192}
+    scoped_provider_binding = (
+        "NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}"
+    )
+    assert workflow.count(scoped_provider_binding) == 2
+    assert "secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY" not in workflow
     free_models = generated_config["provider"]["opencode-free"]["models"]
     paid_zen_models = generated_config["provider"]["opencode"]["models"]
     assert set(free_models) == {
@@ -1647,11 +1652,20 @@ def test_opencode_excludes_queue_self_check_from_every_failed_check_path():
         {"name": "scan-pr-queue", "conclusion": "FAILURE"},
         {"name": "real-peer-check", "conclusion": "FAILURE"},
     ]
-    retained = [
-        check
-        for check in fixtures
-        if check.get("name", "") != "scan-pr-queue"
-    ]
+    extracted_filter = re.search(
+        rf"^\s+\|\s+({re.escape(unconditional_filter)})$",
+        workflow,
+        re.MULTILINE,
+    )
+    assert extracted_filter is not None
+    jq_result = subprocess.run(
+        ["jq", "-c", f"[.[] | {extracted_filter.group(1)}]"],
+        input=json.dumps(fixtures),
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    retained = json.loads(jq_result.stdout)
     assert retained == [{"name": "real-peer-check", "conclusion": "FAILURE"}]
 
 
diff --git a/tests/test_opencode_model_pool_runner.py b/tests/test_opencode_model_pool_runner.py
index fd09efcdf..3769fb42d 100644
--- a/tests/test_opencode_model_pool_runner.py
+++ b/tests/test_opencode_model_pool_runner.py
@@ -808,11 +808,12 @@ def test_nvidia_nim_candidate_requires_key(
     monkeypatch.setenv("NVIDIA_API_KEY", "ambient-provider-key")
     result = run_failed_model(
         tmp_path,
+        extra_env={"NVIDIA_API_KEY": "legacy-provider-key"},
         model_candidates="nvidia-nim/nvidia/nemotron-3-ultra-550b-a55b",
     )
 
     assert result.returncode == 1
-    assert "NVIDIA_API_KEY is not configured" in result.stdout
+    assert "scoped NVIDIA_NIM_API_KEY is not configured" in result.stdout
     assert "attempt 1/1" not in result.stdout
 
 

From 8b58fa080771aa4c6df1c64487a729f893ffeef4 Mon Sep 17 00:00:00 2001
From: Seongho Bae <me@seonghobae.me>
Date: Sat, 1 Aug 2026 05:52:12 +0900
Subject: [PATCH 10/10] test(review): scope NIM secrets to review job

---
 tests/test_opencode_agent_contract.py | 29 ++++++++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)

diff --git a/tests/test_opencode_agent_contract.py b/tests/test_opencode_agent_contract.py
index a1e45a1ab..9e616d5d5 100644
--- a/tests/test_opencode_agent_contract.py
+++ b/tests/test_opencode_agent_contract.py
@@ -201,7 +201,34 @@ def test_opencode_model_pool_sets_high_effort_for_capable_candidates():
     scoped_provider_binding = (
         "NVIDIA_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}"
     )
-    assert workflow.count(scoped_provider_binding) == 2
+    jobs_text = workflow[workflow.index("\njobs:\n") + len("\njobs:\n") :]
+    job_headers = list(
+        re.finditer(r"^  ([A-Za-z0-9_-]+):\n", jobs_text, re.MULTILINE)
+    )
+    job_blocks = {
+        match.group(1): jobs_text[
+            match.start() : (
+                job_headers[index + 1].start()
+                if index + 1 < len(job_headers)
+                else len(jobs_text)
+            )
+        ]
+        for index, match in enumerate(job_headers)
+    }
+    privileged_review_job = job_blocks["opencode-review-target"]
+
+    assert privileged_review_job.count(scoped_provider_binding) == 2
+    assert (
+        privileged_review_job.count(
+            "NVIDIA_NIM_API_KEY: ${{ secrets.NVIDIA_NIM_API_KEY }}"
+        )
+        == 2
+    )
+    for job_name, job_block in job_blocks.items():
+        if job_name == "opencode-review-target":
+            continue
+        assert "secrets.NVIDIA_NIM_API_KEY" not in job_block, job_name
+        assert "secrets.NVIDIA_API_KEY" not in job_block, job_name
     assert "secrets.NVIDIA_NIM_API_KEY || secrets.NVIDIA_API_KEY" not in workflow
     free_models = generated_config["provider"]["opencode-free"]["models"]
     paid_zen_models = generated_config["provider"]["opencode"]["models"]