feat: subagent as task

crates/forge_app/src/agent_executor.rs

@@ -36,12 +36,16 @@ impl<S: Services> AgentExecutor<S> {
     }
 
     /// Executes an agent tool call by creating a new chat request for the
-    /// specified agent.
+    /// Executes an agent tool call by creating a new chat request for the
+    /// specified agent. If conversation_id is provided, the agent will reuse
+    /// that conversation, maintaining context across invocations. Otherwise,
+    /// a new conversation is created.
     pub async fn execute(
         &self,
         agent_id: AgentId,
         task: String,
         ctx: &ToolCallContext,
+        conversation_id: Option<String>,
     ) -> anyhow::Result<ToolOutput> {
         ctx.send_tool_input(
             TitleFormat::debug(format!(
@@ -52,13 +56,23 @@ impl<S: Services> AgentExecutor<S> {
         )
         .await?;
 
-        // Create a new conversation for agent execution
-        let conversation = Conversation::generate().title(task.clone());
-        self.services
-            .conversation_service()
-            .upsert_conversation(conversation.clone())
-            .await?;
-        // Execute the request through the ForgeApp
+        // Reuse existing conversation if provided, otherwise create a new one
+        let conversation = if let Some(cid) = conversation_id {
+            let conversation_id = forge_domain::ConversationId::parse(&cid)
+                .map_err(|_| Error::ConversationNotFound { id: cid.clone() })?;
+            self.services
+                .conversation_service()
+                .find_conversation(&conversation_id)
+                .await?
+                .ok_or(Error::ConversationNotFound { id: cid })?
+        } else {
+            let conversation = Conversation::generate().title(task.clone());
+            self.services
+                .conversation_service()
+                .upsert_conversation(conversation.clone())
+                .await?;
+            conversation
+        };
         let app = crate::ForgeApp::new(self.services.clone());
         let mut response_stream = app
             .chat(

crates/forge_app/src/dto/anthropic/transforms/capitalize_tool_names.rs

@@ -21,6 +21,7 @@ impl Transformer for CapitalizeToolNames {
             tool.name = match tool.name.as_str() {
                 "read" => "Read".to_string(),
                 "write" => "Write".to_string(),
+                "task" => "Task".to_string(),
                 _ => tool.name.clone(),
             };
         }

crates/forge_app/src/error.rs

@@ -40,6 +40,9 @@ pub enum Error {
     #[error("Agent '{0}' not found")]
     AgentNotFound(forge_domain::AgentId),
 
+    #[error("Conversation '{id}' not found")]
+    ConversationNotFound { id: String },
+
     #[error("No active provider configured")]
     NoActiveProvider,
 

crates/forge_app/src/fmt/fmt_input.rs

@@ -125,6 +125,9 @@ impl FormatContent for ToolCatalog {
                     .into(),
             ),
             ToolCatalog::TodoRead(_) => Some(TitleFormat::debug("Read Todos").into()),
+            ToolCatalog::Task(input) => {
+                Some(TitleFormat::debug("Task").sub_title(&input.agent_id).into())
+            }
         }
     }
 }

crates/forge_app/src/orch.rs

@@ -6,6 +6,7 @@ use async_recursion::async_recursion;
 use derive_setters::Setters;
 use forge_domain::{Agent, *};
 use forge_template::Element;
+use futures::future::join_all;
 use tokio::sync::Notify;
 use tracing::warn;
 
@@ -53,27 +54,57 @@ impl<S: AgentService> Orchestrator<S> {
 
     // Helper function to get all tool results from a vector of tool calls
     #[async_recursion]
-    async fn execute_tool_calls<'a>(
+    async fn execute_tool_calls(
         &mut self,
         tool_calls: &[ToolCallFull],
         tool_context: &ToolCallContext,
     ) -> anyhow::Result<Vec<(ToolCallFull, ToolResult)>> {
-        // Always process tool calls sequentially
-        let mut tool_call_records = Vec::with_capacity(tool_calls.len());
+        let task_tool_name = ToolKind::Task.name();
+
+        // Use a case-insensitive comparison since the model may send "Task" or "task".
+        let is_task = |tc: &ToolCallFull| {
+            tc.name
+                .as_str()
+                .eq_ignore_ascii_case(task_tool_name.as_str())
+        };
+
+        // Partition into task tool calls (run in parallel) and all others (run
+        // sequentially). Use a case-insensitive comparison since the model may
+        // send "Task" or "task".
+        let is_task_call =
+            |tc: &&ToolCallFull| tc.name.as_str().to_lowercase() == task_tool_name.as_str();
+        let (task_calls, other_calls): (Vec<_>, Vec<_>) = tool_calls.iter().partition(is_task_call);
+
+        // Execute task tool calls in parallel — mirrors how direct agent-as-tool calls
+        // work.
+        let task_results: Vec<(ToolCallFull, ToolResult)> = join_all(
+            task_calls
+                .iter()
+                .map(|tc| self.services.call(&self.agent, tool_context, (*tc).clone())),
+        )
+        .await
+        .into_iter()
+        .zip(task_calls.iter())
+        .map(|(result, tc)| ((*tc).clone(), result))
+        .collect();
 
         let system_tools = self
             .tool_definitions
             .iter()
             .map(|tool| &tool.name)
             .collect::<HashSet<_>>();
 
-        for tool_call in tool_calls {
+        // Process non-task tool calls sequentially (preserving UI notifier handshake
+        // and hooks).
+        let mut other_results: Vec<(ToolCallFull, ToolResult)> =
+            Vec::with_capacity(other_calls.len());
+        for tool_call in &other_calls {
             // Send the start notification for system tools and not agent as a tool
             let is_system_tool = system_tools.contains(&tool_call.name);
             if is_system_tool {
                 let notifier = Arc::new(Notify::new());
                 self.send(ChatResponse::ToolCallStart {
-                    tool_call: tool_call.clone(),
+                    tool_call: (*tool_call).clone(),
                     notifier: notifier.clone(),
                 })
                 .await?;
@@ -87,7 +118,7 @@ impl<S: AgentService> Orchestrator<S> {
             let toolcall_start_event = LifecycleEvent::ToolcallStart(EventData::new(
                 self.agent.clone(),
                 self.agent.model.clone(),
-                ToolcallStartPayload::new(tool_call.clone()),
+                ToolcallStartPayload::new((*tool_call).clone()),
             ));
             self.hook
                 .handle(&toolcall_start_event, &mut self.conversation)
@@ -96,14 +127,14 @@ impl<S: AgentService> Orchestrator<S> {
             // Execute the tool
             let tool_result = self
                 .services
-                .call(&self.agent, tool_context, tool_call.clone())
+                .call(&self.agent, tool_context, (*tool_call).clone())
                 .await;
 
             // Fire the ToolcallEnd lifecycle event (fires on both success and failure)
             let toolcall_end_event = LifecycleEvent::ToolcallEnd(EventData::new(
                 self.agent.clone(),
                 self.agent.model.clone(),
-                ToolcallEndPayload::new(tool_call.clone(), tool_result.clone()),
+                ToolcallEndPayload::new((*tool_call).clone(), tool_result.clone()),
             ));
             self.hook
                 .handle(&toolcall_end_event, &mut self.conversation)
@@ -114,11 +145,23 @@ impl<S: AgentService> Orchestrator<S> {
                 self.send(ChatResponse::ToolCallEnd(tool_result.clone()))
                     .await?;
             }
-            // Ensure all tool calls and results are recorded
-            // Adding task completion records is critical for compaction to work correctly
-            tool_call_records.push((tool_call.clone(), tool_result));
+            other_results.push(((*tool_call).clone(), tool_result));
         }
 
+        // Reconstruct results in the original order of tool_calls.
+        let mut task_iter = task_results.into_iter();
+        let mut other_iter = other_results.into_iter();
+        let tool_call_records = tool_calls
+            .iter()
+            .map(|tc| {
+                if is_task(tc) {
+                    task_iter.next().expect("task result count mismatch")
+                } else {
+                    other_iter.next().expect("other result count mismatch")
+                }
+            })
+            .collect();
+
         Ok(tool_call_records)
     }
 

crates/forge_app/src/snapshots/forge_app__tool_registry__all_rendered_tool_descriptions.snap

@@ -391,3 +391,75 @@ Retrieves the current todo list for this coding session. Use this tool to check
 ## Output
 
 Returns all current todos with their IDs, content, and status (`pending`, `in_progress`, `completed`). If no todos exist yet, returns an empty list.
+
+---
+
+### task
+
+Launch a new agent to handle complex, multi-step tasks autonomously. 
+
+The task tool launches specialized agents (subprocesses) that autonomously handle complex tasks. Each agent type has specific capabilities and tools available to it.
+
+Available agent types and the tools they have access to:
+- **sage**: Specialized in researching codebases
+  - Tools: read, fs_search, sem_search, fetch
+- **debug**: Specialized in debugging issues
+  - Tools: read, shell, fs_search, sem_search, fetch
+
+When using the task tool, you must specify a agent_id parameter to select which agent type to use.
+
+When NOT to use the task tool:
+- If you want to read a specific file path, use the read or fs_search tool instead of the task tool, to find the match more quickly
+- If you are searching for a specific class definition like "class Foo", use the fs_search tool instead, to find the match more quickly
+- If you are searching for code within a specific file or set of 2-3 files, use the read tool instead of the task tool, to find the match more quickly
+- Other tasks that are not related to the agent descriptions above
+
+
+Usage notes:
+- Always include a short description (3-5 words) summarizing what the agent will do
+- Launch multiple agents concurrently whenever possible, to maximize performance; to do that, use a single message with multiple tool uses
+- When the agent is done, it will return a single message back to you. The result returned by the agent is not visible to the user. To show the user the result, you should send a text message back to the user with a concise summary of the result.
+- Agents can be resumed using the \`session_id\` parameter by passing the agent ID from a previous invocation. When resumed, the agent continues with its full previous context preserved. When NOT resuming, each invocation starts fresh and you should provide a detailed task description with all necessary context.
+- When the agent is done, it will return a single message back to you along with its agent ID. You can use this ID to resume the agent later if needed for follow-up work.
+- Provide clear, detailed prompts so the agent can work autonomously and return exactly the information you need.
+- Agents with "access to current context" can see the full conversation history before the tool call. When using these agents, you can write concise prompts that reference earlier context (e.g., "investigate the error discussed above") instead of repeating information. The agent will receive all prior messages and understand the context.
+- The agent's outputs should generally be trusted
+- Clearly tell the agent whether you expect it to write code or just to do research (search, file reads, web fetches, etc.), since it is not aware of the user's intent
+- If the agent description mentions that it should be used proactively, then you should try your best to use it without the user having to ask for it first. Use your judgement.
+- If the user specifies that they want you to run agents "in parallel", you MUST send a single message with multiple task tool use content blocks. For example, if you need to launch both a build-validator agent and a test-runner agent in parallel, send a single message with both tool calls.
+
+Example usage:
+
+<example_agent_descriptions>
+"test-runner": use this agent after you are done writing code to run tests
+"greeting-responder": use this agent when to respond to user greetings with a friendly joke
+</example_agent_description>
+
+<example>
+user: "Please write a function that checks if a number is prime"
+assistant: Sure let me write a function that checks if a number is prime
+assistant: First let me use the write tool to write a function that checks if a number is prime
+assistant: I'm going to use the write tool to write the following code:
+<code>
+function isPrime(n) {
+  if (n <= 1) return false
+  for (let i = 2; i * i <= n; i++) {
+    if (n % i === 0) return false
+  }
+  return true
+}
+</code>
+<commentary>
+Since a significant piece of code was written and the task was completed, now use the test-runner agent to run the tests
+</commentary>
+assistant: Now let me use the test-runner agent to run the tests
+assistant: Uses the task tool to launch the test-runner agent
+</example>
+
+<example>
+user: "Hello"
+<commentary>
+Since the user is greeting, use the greeting-responder agent to respond with a friendly joke
+</commentary>
+assistant: "I'm going to use the task tool to launch the greeting-responder agent"
+</example>

crates/forge_app/src/system_prompt.rs

@@ -110,6 +110,8 @@ impl<S: SkillFetchService + ShellService> SystemPrompt<S> {
                 model: None,
                 tool_names,
                 extensions,
+                agents: Vec::new(), /* Empty for system prompt (agents list is for tool
+                                     * descriptions only) */
             };
 
             let static_block = TemplateEngine::default()

crates/forge_app/src/tool_executor.rs

@@ -315,6 +315,10 @@ impl<
                 let todos = context.get_todos()?;
                 ToolOperation::TodoRead { output: todos }
             }
+            ToolCatalog::Task(_) => {
+                // Task tools are handled in ToolRegistry before reaching here
+                unreachable!("Task tool should be handled in ToolRegistry")
+            }
         })
     }