fix: enable parallel tool calls by default (#2111)

Author: amitksingh1490

benchmarks/README.md

@@ -4,6 +4,22 @@ A flexible evaluation framework for running automated tests and benchmarks again
 
 ## Quick Start
 
+### Setup
+
+Before running evaluations, create a `forgee` symlink to the debug binary:
+
+```bash
+# Create symlink in your PATH (e.g., ~/bin or /usr/local/bin)
+ln -sf /path/to/code-forge/target/debug/forge ~/forgee
+
+# Or if ~/bin is in your PATH
+ln -sf $(pwd)/target/debug/forge ~/bin/forgee
+```
+
+**Why is this needed?** Tasks execute in temporary directories, so relative paths like `../../target/debug/forge` won't work. The `forgee` symlink provides a stable absolute path that works from any directory.
+
+### Running Evaluations
+
 ```bash
 # Run an evaluation
 npm run eval ./evals/create_skill/task.yml
@@ -58,12 +74,12 @@ before_run:
 
 # Required: Command(s) to execute for each test case
 # Single command
-run: ../../target/debug/forge -p '{{prompt}}'
+run: forgee -p '{{prompt}}'
 
 # Or multiple commands (executed sequentially)
 run:
   - echo "Step 1: {{task}}"
-  - ../../target/debug/forge -p '{{prompt}}'
+  - forgee -p '{{prompt}}'
   - echo "Step 2: Complete"
 
 # Execution configuration

benchmarks/evals/create_skill/task.yml

@@ -1,12 +1,11 @@
-before_run:
-  - cargo build
-run: FORGE_DEBUG_REQUESTS='{{tmp_dir}}/context.json' ../../target/debug/forge -p '{{skill}}'
+# No before_run needed - binary should be pre-built
+run: FORGE_DEBUG_REQUESTS='{{dir}}/context.json' forgee -p '{{skill}}'
 parallelism: 10
 timeout: 60
 early_exit: true
 validations:
   - name: "Uses create-skill tool"
     type: shell
-    command: "jq -e '[.messages[]?.tool_calls[]? | select(.function.name == \"skill\") | .function.arguments | fromjson | .name == \"create-skill\"] | any' {{tmp_dir}}/context.json"
+    command: "jq -e '[.messages[]?.tool_calls[]? | select(.function.name == \"skill\") | .function.arguments | fromjson | .name == \"create-skill\"] | any' {{dir}}/context.json"
 sources:
   - csv: create_skill_tasks.csv

benchmarks/evals/parallel_tool_calls/.gitignore

@@ -0,0 +1 @@
+debug/

benchmarks/evals/parallel_tool_calls/parallel_tool_calls_tasks.csv

@@ -0,0 +1,8 @@
+task
+"Read the contents of AGENTS.md, README.md, and Cargo.toml files"
+"Show me AGENTS.md, forge.schema.json, and rust-toolchain.toml files"
+"I need to see Cargo.toml, LICENSE, and README.md files"
+"Display the contents of forge.schema.json, AGENTS.md, and Cargo.toml simultaneously"
+"Read LICENSE, README.md, and rust-toolchain.toml in parallel"
+"Show me the files: AGENTS.md, Cargo.toml, forge.schema.json, and LICENSE"
+"I want to see README.md, rust-toolchain.toml, and forge.schema.json at once"

benchmarks/evals/parallel_tool_calls/task.yml

@@ -0,0 +1,18 @@
+# No before_run needed - binary should be pre-built
+run:
+  - # Clone into `tmp/task` dir
+  - git clone --depth=1 --branch main https://github.com/antinomyhq/forge .
+  - FORGE_OVERRIDE_PROVIDER=open_router FORGE_OVERRIDE_MODEL={{model}} FORGE_DEBUG_REQUESTS='{{dir}}/context.json' forgee -p '{{task}}'
+parallelism: 5
+timeout: 120
+early_exit: true
+validations:
+  - name: "Uses parallel tool calls"
+    type: shell
+    command: |
+      # Check if there's at least one message with 2 or more tool calls executed in parallel
+      jq -e '[.messages[]?.tool_calls? | select(. != null and length >= 2)] | length > 0' {{dir}}/context.json
+sources:
+  - value:
+      - model: "openai/gpt-5.2"
+  - csv: parallel_tool_calls_tasks.csv

crates/forge_app/src/dto/openai/request.rs

@@ -365,7 +365,8 @@ impl From<Context> for Request {
             models: Default::default(),
             route: Default::default(),
             provider: Default::default(),
-            parallel_tool_calls: Some(false),
+            parallel_tool_calls: Some(true), /* Default to true, transformers will adjust based
+                                              * on model capabilities */
             stream_options: Some(StreamOptions { include_usage: Some(true) }),
             session_id: context.conversation_id.map(|id| id.to_string()),
             reasoning: context.reasoning,