fix: harden ensure_fresh_token_all + add --timeout-multiplier flag

- _common.sh: ensure_fresh_token_all now auto-discovers accounts if
  called before setup_multi_accounts (prevents silent no-op refresh)
- k8s_docs_2config.sh, linuxflbench_2config.sh, swebenchpro_2config.sh:
  add --timeout-multiplier CLI flag and make TIMEOUT_MULTIPLIER
  overridable via env var

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

Author: LoCoBench Bot

configs/_common.sh

@@ -392,6 +392,17 @@ setup_dual_accounts() { setup_multi_accounts; }
 
 # Refresh tokens for all registered accounts.
 ensure_fresh_token_all() {
+    # Some scripts call this before setup_multi_accounts/setup_dual_accounts.
+    # Ensure account homes are discovered so refresh is never silently skipped.
+    if [ ${#CLAUDE_HOMES[@]} -eq 0 ]; then
+        setup_multi_accounts
+    fi
+
+    # Safety fallback in case setup_multi_accounts did not populate for any reason.
+    if [ ${#CLAUDE_HOMES[@]} -eq 0 ]; then
+        CLAUDE_HOMES=("${HOME}")
+    fi
+
     for home_dir in "${CLAUDE_HOMES[@]}"; do
         echo "Refreshing token for HOME=$home_dir ..."
         HOME="$home_dir" ensure_fresh_token

configs/k8s_docs_2config.sh

@@ -14,6 +14,7 @@
 #   --model MODEL          Override model (default: claude-opus-4-6)
 #   --category CATEGORY    Run category (default: official)
 #   --parallel N           Number of parallel task subshells (default: 1)
+#   --timeout-multiplier N Override Harbor timeout multiplier (default: 3)
 #
 # Prerequisites:
 #   - ~/evals/.env.local with USE_SUBSCRIPTION=true (default: 2-account Max subscription)
@@ -60,7 +61,7 @@ TASKS_DIR="/home/stephanie_jarmak/CodeContextBench/benchmarks/ccb_k8sdocs"
 AGENT_PATH="agents.claude_baseline_agent:BaselineClaudeCodeAgent"
 MODEL="${MODEL:-anthropic/claude-opus-4-6}"
 CONCURRENCY=2
-TIMEOUT_MULTIPLIER=3  # 3x default for 900s task timeout
+TIMEOUT_MULTIPLIER="${TIMEOUT_MULTIPLIER:-3}"  # 3x default for 900s task timeout
 RUN_BASELINE=true
 RUN_FULL=true
 CATEGORY="${CATEGORY:-official}"
@@ -88,6 +89,10 @@ while [[ $# -gt 0 ]]; do
             PARALLEL_JOBS="$2"
             shift 2
             ;;
+        --timeout-multiplier)
+            TIMEOUT_MULTIPLIER="$2"
+            shift 2
+            ;;
         *)
             echo "Unknown option: $1"
             exit 1

configs/linuxflbench_2config.sh

@@ -17,6 +17,7 @@
 #   --full-only            Run only MCP-Full (sourcegraph_full)
 #   --model MODEL          Override model (default: claude-opus-4-6)
 #   --category CATEGORY    Run category (default: official)
+#   --timeout-multiplier N Override Harbor timeout multiplier (default: 10)
 #
 # Prerequisites:
 #   - ~/evals/.env.local with USE_SUBSCRIPTION=true (default: 2-account Max subscription)
@@ -67,7 +68,7 @@ TASKS_DIR="/home/stephanie_jarmak/CodeContextBench/benchmarks/${SUITE}"
 AGENT_PATH="agents.claude_baseline_agent:BaselineClaudeCodeAgent"
 MODEL="${MODEL:-anthropic/claude-opus-4-6}"
 CONCURRENCY=1
-TIMEOUT_MULTIPLIER=10
+TIMEOUT_MULTIPLIER="${TIMEOUT_MULTIPLIER:-10}"
 RUN_BASELINE=true
 RUN_FULL=true
 CATEGORY="${CATEGORY:-official}"
@@ -95,6 +96,10 @@ while [[ $# -gt 0 ]]; do
             PARALLEL_JOBS="$2"
             shift 2
             ;;
+        --timeout-multiplier)
+            TIMEOUT_MULTIPLIER="$2"
+            shift 2
+            ;;
         *)
             echo "Unknown option: $1"
             exit 1
@@ -133,6 +138,11 @@ declare -A TASK_SG_REPO_NAMES=(
     ["lfl-sound-53441"]="github.com/sg-benchmarks/linux--07c4ee00"
 )
 
+# Per-task timeout multiplier overrides for known long-running tasks.
+declare -A TASK_TIMEOUT_MULTIPLIERS=(
+    ["lfl-nfs-117651"]="15"
+)
+
 # Derive short model name for run directory
 _model_lower=$(echo "$MODEL" | awk -F/ '{print $NF}' | tr '[:upper:]' '[:lower:]')
 case "$_model_lower" in
@@ -197,6 +207,7 @@ _linuxflbench_run_single() {
     local jobs_base=${5:-$JOBS_BASE}
     local jobs_subdir="${jobs_base}/${config}"
     local task_path="${TASKS_DIR}/${task_id}"
+    local timeout_mult="${TASK_TIMEOUT_MULTIPLIERS[$task_id]:-$TIMEOUT_MULTIPLIER}"
 
     mkdir -p "$jobs_subdir"
 
@@ -205,7 +216,7 @@ _linuxflbench_run_single() {
         return 1
     fi
 
-    echo "Running task: $task_id ($config) [HOME=$task_home]"
+    echo "Running task: $task_id ($config) [HOME=$task_home] [timeout x${timeout_mult}]"
 
     local sg_repo="${TASK_SG_REPO_NAMES[$task_id]:-}"
     if [ -n "$sg_repo" ]; then
@@ -220,7 +231,7 @@ _linuxflbench_run_single() {
         --model "$MODEL" \
         --jobs-dir "$jobs_subdir" \
         -n $CONCURRENCY \
-        --timeout-multiplier $TIMEOUT_MULTIPLIER \
+        --timeout-multiplier "$timeout_mult" \
         2>&1 | tee "${jobs_subdir}/${task_id}.log" \
         || {
             echo "WARNING: Task $task_id ($config) failed (exit code: $?)"

configs/swebenchpro_2config.sh

@@ -15,6 +15,7 @@
 #   --concurrency N        Number of concurrent tasks (default: 2)
 #   --category CATEGORY    Run category (default: official)
 #   --parallel N           Number of parallel task subshells (default: 1)
+#   --timeout-multiplier N Override Harbor timeout multiplier (default: 10)
 #
 # Prerequisites:
 #   - ~/evals/.env.local with USE_SUBSCRIPTION=true (default: 2-account Max subscription)
@@ -60,7 +61,7 @@ ensure_fresh_token
 AGENT_PATH="agents.claude_baseline_agent:BaselineClaudeCodeAgent"
 MODEL="${MODEL:-anthropic/claude-opus-4-6}"
 CONCURRENCY=2
-TIMEOUT_MULTIPLIER=10
+TIMEOUT_MULTIPLIER="${TIMEOUT_MULTIPLIER:-10}"
 RUN_BASELINE=true
 RUN_FULL=true
 CATEGORY="${CATEGORY:-official}"
@@ -92,6 +93,10 @@ while [[ $# -gt 0 ]]; do
             PARALLEL_JOBS="$2"
             shift 2
             ;;
+        --timeout-multiplier)
+            TIMEOUT_MULTIPLIER="$2"
+            shift 2
+            ;;
         *)
             echo "Unknown option: $1"
             exit 1