agents: fix vllm crash on malformed tool args, always use API

Three fixes:

1. Sanitize tool call arguments before pushing to conversation
   history — vllm re-parses them as JSON on the next request and
   crashes on invalid JSON from a previous turn. Malformed args now
   get replaced with {} and the model gets an error message telling
   it to retry with valid JSON.

2. Remove is_split special case — split goes through the normal
   job_consolidation_agent path like all other agents.

3. call_for_def always uses API when api_base_url is configured,
   regardless of tools field. Remove tools field from all .agent
   files — memory tools are always provided by the API layer.

Also adds prompt size guard (800KB max) to catch oversized prompts
before they hit the model context limit.

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

poc-memory/src/agents/api.rs

@@ -70,8 +70,21 @@ pub async fn call_api_with_tools(
         let has_tools = msg.tool_calls.as_ref().is_some_and(|tc| !tc.is_empty());
 
         if has_tools {
-            // Push the assistant message with tool calls
+            // Push the assistant message with tool calls.
-            messages.push(msg.clone());
+            // Sanitize arguments: vllm re-parses them as JSON when
+            // preprocessing the conversation, so invalid JSON from the
+            // model crashes the next request.
+            let mut sanitized = msg.clone();
+            if let Some(ref mut calls) = sanitized.tool_calls {
+                for call in calls {
+                    if serde_json::from_str::<serde_json::Value>(&call.function.arguments).is_err() {
+                        log(&format!("sanitizing malformed args for {}: {}",
+                            call.function.name, &call.function.arguments));
+                        call.function.arguments = "{}".to_string();
+                    }
+                }
+            }
+            messages.push(sanitized);
 
             // Execute each tool call
             for call in msg.tool_calls.as_ref().unwrap() {
@@ -79,8 +92,17 @@ pub async fn call_api_with_tools(
                     call.function.name,
                     &call.function.arguments));
 
-                let args: serde_json::Value = serde_json::from_str(&call.function.arguments)
+                let args: serde_json::Value = match serde_json::from_str(&call.function.arguments) {
-                    .unwrap_or_default();
+                    Ok(v) => v,
+                    Err(_) => {
+                        log(&format!("malformed tool call args: {}", &call.function.arguments));
+                        messages.push(Message::tool_result(
+                            &call.id,
+                            "Error: your tool call had malformed JSON arguments. Please retry with valid JSON.",
+                        ));
+                        continue;
+                    }
+                };
 
                 let output = if call.function.name.starts_with("memory_") {
                     let prov = format!("agent:{}", agent);