follow master change, use gpt-5.4-mini max_tokens=1024

server-rs/Cargo.lock

@@ -4447,6 +4447,7 @@ name = "platform-editor-agent"
 version = "0.1.0"
 dependencies = [
  "hmac",
+ "platform-agent",
  "platform-llm",
  "serde",
  "serde_json",

server-rs/crates/api-server/src/editor_agent/api.rs

@@ -167,9 +167,9 @@ pub async fn editor_agent_message(
     let tool_context = context::build_tool_context(&document);
 
     // Build and run agent
-    let llm_client = state.llm_client().ok_or_else(|| {
+    let llm_client = state.creative_agent_gpt5_client().ok_or_else(|| {
         AppError::from_status(axum::http::StatusCode::SERVICE_UNAVAILABLE)
-            .with_details(json!({ "message": "LLM client not configured" }))
+            .with_details(json!({ "message": "Creative Agent GPT-5 client not configured" }))
     })?;
     let llm_client = llm_client.clone();
     let pricing = state.editor_generation_pricing().await.map_err(|error| {

server-rs/crates/platform-editor-agent/Cargo.toml

@@ -6,6 +6,7 @@ license.workspace = true
 
 [dependencies]
 hmac = { workspace = true }
+platform-agent = { workspace = true }
 platform-llm = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }

server-rs/crates/platform-editor-agent/src/agent/agent.rs

@@ -4,9 +4,13 @@ use crate::framework::error::PromptError;
 use crate::framework::hook::Hook;
 use crate::framework::memory::AgentMemory;
 use crate::framework::tool::{Tool, ToolDyn};
-use platform_llm::{LlmClient, LlmMessage};
+use platform_agent::CREATIVE_AGENT_GPT5_MODEL;
+use platform_llm::{LlmClient, LlmMessage, LlmTextRequest};
 use serde_json::Value;
 
+const EDITOR_AGENT_LLM_MAX_OUTPUT_TOKENS: u32 = 1024;
+const EDITOR_AGENT_LLM_REQUEST_TIMEOUT_MS: u64 = 60_000;
+
 pub struct LlmCompletionModel {
     client: LlmClient,
 }
@@ -16,9 +20,7 @@ impl LlmApiAdaptor<LlmMessage> for LlmCompletionModel {
         &self,
         messages: impl Iterator<Item = &'a LlmMessage> + Send,
     ) -> Result<String, PromptError> {
-        use platform_llm::LlmTextRequest;
-        let request =
-            LlmTextRequest::new(messages.cloned().collect()).with_request_timeout_ms(30_000);
+        let request = build_editor_agent_llm_request(messages.cloned().collect());
         let response = self
             .client
             .request_text(request)
@@ -36,6 +38,13 @@ impl LlmApiAdaptor<LlmMessage> for LlmCompletionModel {
     }
 }
 
+fn build_editor_agent_llm_request(messages: Vec<LlmMessage>) -> LlmTextRequest {
+    LlmTextRequest::new(messages)
+        .with_model(CREATIVE_AGENT_GPT5_MODEL)
+        .with_max_tokens(EDITOR_AGENT_LLM_MAX_OUTPUT_TOKENS)
+        .with_request_timeout_ms(EDITOR_AGENT_LLM_REQUEST_TIMEOUT_MS)
+}
+
 pub struct LlmChatAgentBuilder {
     client: Option<LlmClient>,
     system_prompt_parts: Vec<String>,
@@ -172,3 +181,24 @@ fn build_tools_system_prompt(base_prompt: &str, tool_specs: &[ToolPromptSpec]) -
 
     prompt
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn editor_agent_llm_request_keeps_gpt5_contract() {
+        let request = build_editor_agent_llm_request(vec![
+            LlmMessage::system("系统提示"),
+            LlmMessage::user("用户请求"),
+        ]);
+
+        assert_eq!(
+            request.model.as_deref(),
+            Some(platform_agent::CREATIVE_AGENT_GPT5_MODEL)
+        );
+        assert_eq!(request.max_tokens, Some(1024));
+        assert_eq!(request.request_timeout_ms, Some(60_000));
+        assert_eq!(request.messages.len(), 2);
+    }
+}