feat: add Qwen3.5 model catalog and agentic evaluation framework

Models:
- configs/models.conf: catalog with Qwen3.5-35B-A3B (MoE, top pick),
  Qwen3.5-27B (dense), Qwen3-Coder-30B-A3B (agentic/coding)
- Updated benchmark setup to show catalog with download status
- docs/model-recommendations.md: memory planning, quantization guide

Agentic evaluation:
- scripts/agentic/setup.sh: installs inspect-ai, evalplus, bigcodebench
  in a Python venv
- scripts/agentic/run-eval.sh: runs evaluations against local LLM server
  (ollama or llama.cpp). Suites: quick (HumanEval+IFEval), code
  (EvalPlus+BigCodeBench), tooluse (BFCL), full (all)
- bin/agentic: dispatcher with help
- docs/agentic-benchmarks.md: methodology, framework comparison, model
  recommendations for agentic use

Updated: Makefile (6 new targets), README, CLAUDE.md, docs/references.md

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

scripts/benchmark/setup.sh

@@ -8,13 +8,13 @@ source "$SCRIPT_DIR/../../lib/detect.sh"
 
 TOOLBOXES_REPO="/data/workspace/projects/HomeLab/strix-halo-toolboxes/amd-strix-halo-llamacpp-toolboxes"
 MODEL_DIR="$(data_dir models)"
+MODEL_CATALOG="$PROJECT_ROOT/configs/models.conf"
 
 log_header "Benchmark Setup"
 
 # ── 1. Check toolbox containers ──────────────────────────
 log_info "Checking toolbox containers..."
 
-# Minimum required: vulkan-radv (most stable)
 REQUIRED_TOOLBOXES=("llama-vulkan-radv")
 OPTIONAL_TOOLBOXES=("llama-rocm-6.4.4" "llama-rocm-7.2" "llama-vulkan-amdvlk")
 
@@ -22,7 +22,7 @@ existing=$(detect_toolbox_names 2>/dev/null || true)
 missing=()
 
 for tb in "${REQUIRED_TOOLBOXES[@]}"; do
-    if echo "$existing" | grep -q "^${tb}$"; then
+    if echo "$existing" | grep -qF "$tb"; then
         log_success "Toolbox: $tb"
     else
         missing+=("$tb")
@@ -31,7 +31,7 @@ for tb in "${REQUIRED_TOOLBOXES[@]}"; do
 done
 
 for tb in "${OPTIONAL_TOOLBOXES[@]}"; do
-    if echo "$existing" | grep -q "^${tb}$"; then
+    if echo "$existing" | grep -qF "$tb"; then
         log_success "Toolbox: $tb (optional)"
     else
         log_info "Toolbox not present: $tb (optional)"
@@ -80,26 +80,54 @@ if (( model_count > 0 )); then
     done
 else
     log_warn "No GGUF models found in $MODEL_DIR"
-    log_info "Download a test model. Example:"
+fi
+
+# ── 4. Show model catalog ───────────────────────────────
+log_header "Model Catalog"
+log_info "Available models (from configs/models.conf):"
+echo ""
+printf "  ${BOLD}%-28s %-10s %-8s %s${RESET}\n" "Name" "Category" "Size" "Description"
+echo "  $(printf '%.0s─' {1..70})"
+while IFS='|' read -r name repo file size_gb category desc; do
+    [[ "$name" =~ ^#.*$ || -z "$name" ]] && continue
+    local_file="$MODEL_DIR/$file"
+    status=" "
+    if [[ -f "$local_file" ]]; then
+        status="*"
+    fi
+    printf "  %s %-27s %-10s %4s GB  %s\n" "$status" "$name" "$category" "$size_gb" "$desc"
+done < "$MODEL_CATALOG"
+echo ""
+echo "  (* = downloaded)"
+echo ""
+
+# ── 5. Offer downloads ──────────────────────────────────
+if is_cmd huggingface-cli; then
+    log_info "Download models with:"
     echo ""
-    echo "  # Small (4B, ~3 GB):"
-    echo "  huggingface-cli download Qwen/Qwen3-4B-GGUF Qwen3-4B-Q4_K_M.gguf \\"
-    echo "    --local-dir $MODEL_DIR"
+    echo "  # Recommended starter set:"
+    echo "  # Smoke test (3 GB):"
+    echo "  huggingface-cli download unsloth/Qwen3-4B-GGUF Qwen3-4B-Q4_K_M.gguf --local-dir $MODEL_DIR"
     echo ""
-    echo "  # Medium (14B, ~9 GB):"
-    echo "  huggingface-cli download Qwen/Qwen3-14B-GGUF Qwen3-14B-Q4_K_M.gguf \\"
-    echo "    --local-dir $MODEL_DIR"
+    echo "  # Top pick — Qwen3.5-35B-A3B MoE Q8 (37 GB, ~85 t/s gen):"
+    echo "  huggingface-cli download unsloth/Qwen3.5-35B-A3B-GGUF Qwen3.5-35B-A3B-Q8_0.gguf --local-dir $MODEL_DIR"
+    echo ""
+    echo "  # Agentic/coding — Qwen3-Coder-30B-A3B (18 GB, best for tool use):"
+    echo "  huggingface-cli download unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf --local-dir $MODEL_DIR"
+    echo ""
+    echo "  # Or download any model from catalog:"
+    echo "  # huggingface-cli download REPO FILE --local-dir $MODEL_DIR"
     echo ""
 
-    if is_cmd huggingface-cli; then
-        if confirm "Download Qwen3-4B Q4_K_M (~3 GB) as test model?"; then
-            huggingface-cli download Qwen/Qwen3-4B-GGUF Qwen3-4B-Q4_K_M.gguf \
+    if (( model_count == 0 )); then
+        if confirm "Download Qwen3-4B Q4_K_M (~3 GB) as smoke test model?"; then
+            huggingface-cli download unsloth/Qwen3-4B-GGUF Qwen3-4B-Q4_K_M.gguf \
                 --local-dir "$MODEL_DIR"
             log_success "Model downloaded"
         fi
-    else
-        log_info "Install huggingface-cli: pip install huggingface_hub[cli]"
     fi
+else
+    log_info "Install huggingface-cli: pip install huggingface_hub[cli]"
 fi
 
 log_header "Setup Complete"