feat: replace subagent and skills with unified summon extension (#6964)

Signed-off-by: Travis Longwell <travis@block.xyz>

scripts/test-subrecipes-examples/project_analyzer_parallel.yaml

@@ -17,6 +17,7 @@ prompt: |
   Run two subrecipes in parallel:
     - use file_stats subrecipe to gather file statistics for {{ target_directory }}
     - use code_patterns subrecipe to analyze code patterns in {{ target_directory }}
+  Iteratively `sleep 10` until the delegates complete, then load their output and confirm success
 extensions:
   - type: builtin
     name: developer

scripts/test_providers.sh

@@ -21,8 +21,10 @@ done
 # These are typically preview/experimental models with inconsistent tool-calling behavior.
 # Failures are still reported but don't block PRs.
 ALLOWED_FAILURES=(
+  "google:gemini-2.5-flash"
   "google:gemini-3-pro-preview"
   "openrouter:nvidia/nemotron-3-nano-30b-a3b"
+  "openai:gpt-3.5-turbo"
 )
 
 # Agentic providers handle tools internally and return text results.

scripts/test_subrecipes.sh

@@ -22,7 +22,7 @@ export PATH="$SCRIPT_DIR/target/debug:$PATH"
 # Set default provider and model if not already set
 # Use fast model for CI to speed up tests
 export GOOSE_PROVIDER="${GOOSE_PROVIDER:-anthropic}"
-export GOOSE_MODEL="${GOOSE_MODEL:-claude-3-5-haiku-20241022}"
+export GOOSE_MODEL="${GOOSE_MODEL:-claude-haiku-4-5}"
 
 echo "Using provider: $GOOSE_PROVIDER"
 echo "Using model: $GOOSE_MODEL"
@@ -77,17 +77,17 @@ check_recipe_output() {
   local tmpfile=$1
   local mode=$2
 
-  # Check for unified subagent tool invocation (new format: "─── subagent |")
-  if grep -q "─── subagent" "$tmpfile"; then
-    echo "✓ SUCCESS: Subagent tool invoked"
-    RESULTS+=("✓ Subagent tool invocation ($mode)")
+  # Check for delegate tool invocation (new format: "─── delegate |")
+  if grep -q "─── delegate" "$tmpfile"; then
+    echo "✓ SUCCESS: Delegate tool invoked"
+    RESULTS+=("✓ Delegate tool invocation ($mode)")
   else
-    echo "✗ FAILED: No evidence of subagent tool invocation"
-    RESULTS+=("✗ Subagent tool invocation ($mode)")
+    echo "✗ FAILED: No evidence of delegate tool invocation"
+    RESULTS+=("✗ Delegate tool invocation ($mode)")
   fi
 
-  # Check that both subrecipes were called (shown as "subrecipe: <name>" in output)
-  if grep -q "subrecipe:.*file_stats\|file_stats.*subrecipe" "$tmpfile" && grep -q "subrecipe:.*code_patterns\|code_patterns.*subrecipe" "$tmpfile"; then
+  # Check that both subrecipes were called (shown as "source: <name>" in delegate output)
+  if grep -q "source:.*file_stats\|source.*file_stats" "$tmpfile" && grep -q "source:.*code_patterns\|source.*code_patterns" "$tmpfile"; then
     echo "✓ SUCCESS: Both subrecipes (file_stats, code_patterns) found in output"
     RESULTS+=("✓ Both subrecipes present ($mode)")
   else