Merge branch 'main' into fixBugss

Author: evanliu048

crates/chat-cli/src/cli/chat/consts.rs

@@ -7,10 +7,10 @@ pub const MAX_CURRENT_WORKING_DIRECTORY_LEN: usize = 256;
 /// Limit to send the number of messages as part of chat.
 pub const MAX_CONVERSATION_STATE_HISTORY_LEN: usize = 250;
 
-pub const MAX_TOOL_RESPONSE_SIZE: usize = 800_000;
+/// Actual service limit is 800_000
+pub const MAX_TOOL_RESPONSE_SIZE: usize = 600_000;
 
-/// TODO: Use this to gracefully handle user message sizes.
-#[allow(dead_code)]
+/// Actual service limit is 600_000
 pub const MAX_USER_MESSAGE_SIZE: usize = 600_000;
 
 /// In tokens

crates/chat-cli/src/cli/chat/conversation_state.rs

@@ -25,6 +25,7 @@ use super::consts::{
     DUMMY_TOOL_NAME,
     MAX_CHARS,
     MAX_CONVERSATION_STATE_HISTORY_LEN,
+    MAX_USER_MESSAGE_SIZE,
 };
 use super::context::ContextManager;
 use super::hooks::{
@@ -50,7 +51,10 @@ use super::tools::{
     ToolOrigin,
     ToolSpec,
 };
-use super::util::serde_value_to_document;
+use super::util::{
+    serde_value_to_document,
+    truncate_safe,
+};
 use crate::api_client::model::{
     AssistantResponseMessage,
     ChatMessage,
@@ -527,6 +531,13 @@ impl ConversationState {
         }
     }
 
+    /// Whether or not it is possible to create a summary out of this conversation state.
+    ///
+    /// Currently only checks if we have enough messages in the history to create a summary out of.
+    pub async fn can_create_summary_request(&mut self) -> bool {
+        self.backend_conversation_state(false, true).await.history.len() >= 2
+    }
+
     /// Returns a [FigConversationState] capable of replacing the history of the current
     /// conversation with a summary generated by the model.
     pub async fn create_summary_request(&mut self, custom_prompt: Option<impl AsRef<str>>) -> FigConversationState {
@@ -633,7 +644,8 @@ impl ConversationState {
                     // something is set.
                     tool_content.push_str("<tool result redacted>");
                 }
-                user.content = UserMessageContent::Prompt { prompt: tool_content };
+                let prompt = truncate_safe(&tool_content, MAX_USER_MESSAGE_SIZE).to_string();
+                user.content = UserMessageContent::Prompt { prompt };
             }
         }
     }

crates/chat-cli/src/cli/chat/mod.rs

@@ -958,26 +958,36 @@ impl ChatContext {
                         // Errors from attempting to send too large of a conversation history. In
                         // this case, attempt to automatically compact the history for the user.
                         crate::api_client::ApiClientError::ContextWindowOverflow => {
-                            let history_too_small = self
-                                .conversation_state
-                                .backend_conversation_state(false, true)
-                                .await
-                                .history
-                                .len()
-                                < 2;
-                            if history_too_small {
-                                print_err!(
-                                    "Your conversation is too large - try reducing the size of
-                                the context being passed",
-                                    err
-                                );
+                            if !self.conversation_state.can_create_summary_request().await {
+                                execute!(
+                                    self.output,
+                                    style::SetForegroundColor(Color::Red),
+                                    style::Print("Your conversation is too large to continue.\n"),
+                                    style::SetForegroundColor(Color::Reset),
+                                    style::Print(format!("• Run {} to analyze your context usage\n", "/usage".green())),
+                                    style::Print(format!(
+                                        "• Run {} to reset your conversation state\n",
+                                        "/clear".green()
+                                    )),
+                                    style::SetAttribute(Attribute::Reset),
+                                    style::Print("\n\n"),
+                                )?;
+                                self.conversation_state.reset_next_user_message();
                                 return Ok(ChatState::PromptUser {
                                     tool_uses: None,
                                     pending_tool_index: None,
                                     skip_printing_tools: false,
                                 });
                             }
 
+                            execute!(
+                                self.output,
+                                style::SetForegroundColor(Color::Yellow),
+                                style::Print("The context window has overflowed, summarizing the history..."),
+                                style::SetAttribute(Attribute::Reset),
+                                style::Print("\n\n"),
+                            )?;
+
                             return Ok(ChatState::CompactHistory {
                                 tool_uses: None,
                                 pending_tool_index: None,
@@ -1223,8 +1233,10 @@ impl ChatContext {
         // Check token usage and display warnings if needed
         if pending_tool_index.is_none() {
             // Only display warnings when not waiting for tool approval
-            if let Err(e) = self.display_char_warnings().await {
-                warn!("Failed to display character limit warnings: {}", e);
+            if self.conversation_state.can_create_summary_request().await {
+                if let Err(e) = self.display_char_warnings().await {
+                    warn!("Failed to display character limit warnings: {}", e);
+                }
             }
         }
 

crates/chat-cli/src/cli/chat/tool_manager.rs

@@ -400,8 +400,8 @@ impl ToolManagerBuilder {
                                 let msg = eyre::eyre!(msg);
                                 let total = loading_servers.len();
                                 queue_incomplete_load_message(complete, total, &msg, &mut output)?;
-                                output.flush()?;
                             }
+                            execute!(output, style::Print("\n"),)?;
                             break;
                         },
                     },

crates/q_cli/src/cli/mod.rs

@@ -200,6 +200,12 @@ pub enum CliRootCommands {
         #[arg(trailing_var_arg = true, allow_hyphen_values = true)]
         args: Vec<String>,
     },
+    /// Model Context Protocol (MCP)
+    Mcp {
+        /// Args for the MCP subcommand (passed through to `qchat mcp …`)
+        #[arg(trailing_var_arg = true, allow_hyphen_values = true)]
+        args: Vec<String>,
+    },
     /// Inline shell completions
     #[command(subcommand)]
     Inline(inline::InlineSubcommand),
@@ -235,6 +241,7 @@ impl CliRootCommands {
             CliRootCommands::Version { .. } => "version",
             CliRootCommands::Dashboard => "dashboard",
             CliRootCommands::Chat { .. } => "chat",
+            CliRootCommands::Mcp { .. } => "mcp",
             CliRootCommands::Inline(_) => "inline",
         }
     }
@@ -344,15 +351,16 @@ impl Cli {
                 CliRootCommands::Telemetry(subcommand) => subcommand.execute().await,
                 CliRootCommands::Version { changelog } => Self::print_version(changelog),
                 CliRootCommands::Dashboard => launch_dashboard(false).await,
-                CliRootCommands::Chat { args } => Self::execute_chat(Some(args), true).await,
+                CliRootCommands::Chat { args } => Self::execute_chat("chat", Some(args), true).await,
+                CliRootCommands::Mcp { args } => Self::execute_chat("mcp", Some(args), true).await,
                 CliRootCommands::Inline(subcommand) => subcommand.execute(&cli_context).await,
             },
             // Root command
-            None => Self::execute_chat(None, true).await,
+            None => Self::execute_chat("chat", None, true).await,
         }
     }
 
-    pub async fn execute_chat(args: Option<Vec<String>>, enforce_login: bool) -> Result<ExitCode> {
+    pub async fn execute_chat(subcmd: &str, args: Option<Vec<String>>, enforce_login: bool) -> Result<ExitCode> {
         if enforce_login {
             assert_logged_in().await?;
         }
@@ -369,8 +377,7 @@ impl Cli {
         }
 
         let mut cmd = tokio::process::Command::new(home_local_bin()?.join(CHAT_BINARY_NAME));
-        cmd.arg("chat");
-
+        cmd.arg(subcmd);
         if let Some(args) = args {
             cmd.args(args);
         }

crates/q_cli/src/main.rs

@@ -43,7 +43,7 @@ fn main() -> Result<ExitCode> {
 
     // Hack as clap doesn't expose a custom command help.
     if subcommand.as_deref() == Some("chat") && args.any(|arg| ["--help", "-h"].contains(&arg.as_str())) {
-        runtime.block_on(cli::Cli::execute_chat(Some(vec!["--help".to_owned()]), true))?;
+        runtime.block_on(cli::Cli::execute_chat("chat", Some(vec!["--help".to_owned()]), true))?;
     }
 
     let parsed = match cli::Cli::try_parse() {