fix: trace audit — fix 3 broken verifiers, MANIFEST model fallback, drop 2 TAC tasks

- navidrome test.sh: pytest → go test (Go/Ginkgo project)
- nodebb-notif/nodebb-plugin test.sh: pytest → npx mocha + Mocha reward parsing
- openlibrary Dockerfile.sg_only: pre-install Node.js 22 + Claude Code (sweap-images Node 16 broken)
- generate_manifest.py: model extraction falls back to result.json when config.json missing
  (fixes ccb_feature/ccb_refactor incorrectly showing opus instead of haiku)
- Drop 2 llamacpp TAC tasks (need external RocketChat server, incompatible with benchmark)
- selected_benchmark_tasks.json: 414 → 412 tasks, ccb_test 20 → 18
- Add handoff doc for rerun setup (local Docker + Daytona)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sjarmak

benchmarks/ccb_fix/navidrome-windows-log-fix-001/tests/test.sh

@@ -112,11 +112,11 @@ else
     echo "WARNING: Test patch failed to apply."
 fi
 
-echo "Running specific pytest tests..."
-echo "Tests: "TestLog""
+echo "Running Go tests..."
+echo "Tests: TestLog"
 
-# Run only the specific tests from fail_to_pass and pass_to_pass
-python -m pytest "TestLog" -v 2>&1 | tee test_output.log
+# Navidrome is a Go project using Ginkgo. Run the log package tests.
+go test -v -run "TestLog" ./log/... 2>&1 | tee test_output.log
 TEST_EXIT_CODE=$?
 
 # Write reward for Harbor with partial credit

benchmarks/ccb_fix/nodebb-notif-dropdown-fix-001/tests/test.sh

@@ -6,6 +6,14 @@ FROM jefzda/sweap-images:internetarchive.openlibrary-internetarchive__openlibrar
 
 ENV SOURCEGRAPH_REPO_NAME=sg-evals/openlibrary--92db3454
 
+# Pre-install Node.js 22 + Claude Code at build time.
+# Harbor's install.sh fails on this image (NodeSource GPG needs /dev/tty)
+# but if claude is already on PATH, /tmp/claude_run.sh finds it.
+RUN curl -fsSL https://nodejs.org/dist/v22.14.0/node-v22.14.0-linux-x64.tar.gz \
+    | tar -xz -C /usr/local --strip-components=1 \
+    && node --version && npm --version \
+    && npm install -g @anthropic-ai/claude-code@latest \
+    && which claude && claude --version
 
 RUN curl -LsSf https://astral.sh/uv/0.7.13/install.sh | sh || true
 RUN mkdir -p /logs /workspace