Merge branch 'main' into persona-enablement

Author: dingfeli

build-config/buildspec-macos.yml

@@ -24,7 +24,7 @@ phases:
       - mise install
       - eval "$(mise activate zsh --shims)"
       # Install python deps
-      - python3 -m venv scripts/.env
+      - python3 -m venv build-scripts/.env
       - source build-scripts/.env/bin/activate
       - pip3 install -r build-scripts/requirements.txt
   build:

build-scripts/signing.py

@@ -298,7 +298,9 @@ def apple_notarize_file(file: pathlib.Path, signing_data: CdSigningData):
 
 def get_secretmanager_json(secret_id: str):
     info(f"Loading secretmanager value: {secret_id}")
-    secret_value = run_cmd_output(["aws", "secretsmanager", "get-secret-value", "--secret-id", secret_id])
+    secret_value = run_cmd_output(
+        ["aws", "--region", REGION, "secretsmanager", "get-secret-value", "--secret-id", secret_id]
+    )
     secret_string = json.loads(secret_value)["SecretString"]
     return json.loads(secret_string)
 

crates/chat-cli/src/auth/builder_id.rs

@@ -458,6 +458,12 @@ impl BuilderIdToken {
             Some(_) => TokenType::IamIdentityCenter,
         }
     }
+
+    /// Check if the token is for the internal amzn start URL (`https://amzn.awsapps.com/start`),
+    /// this implies the user will use midway for private specs
+    pub fn is_amzn_user(&self) -> bool {
+        matches!(&self.start_url, Some(url) if url == AMZN_START_URL)
+    }
 }
 
 pub enum PollCreateToken {

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

@@ -21,5 +21,8 @@ pub(crate) const CLIENT_TYPE: &str = "public";
 // The start URL for public builder ID users
 pub const START_URL: &str = "https://view.awsapps.com/start";
 
+// The start URL for internal amzn users
+pub const AMZN_START_URL: &str = "https://amzn.awsapps.com/start";
+
 pub(crate) const DEVICE_GRANT_TYPE: &str = "urn:ietf:params:oauth:grant-type:device_code";
 pub(crate) const REFRESH_GRANT_TYPE: &str = "refresh_token";

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

@@ -35,6 +35,6 @@ pub struct CompactArgs {
 
 impl CompactArgs {
     pub async fn execute(self, os: &Os, session: &mut ChatSession) -> Result<ChatState, ChatError> {
-        session.compact_history(os, self.prompt, self.show_summary).await
+        session.compact_history(os, self.prompt, self.show_summary, true).await
     }
 }

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

@@ -56,7 +56,11 @@ pub enum ContextSubcommand {
         paths: Vec<String>,
     },
     /// Remove specified rules from current profile
-    Remove { paths: Vec<String> },
+    #[command(alias = "rm")]
+    Remove {
+        #[arg(required = true)]
+        paths: Vec<String>,
+    },
     /// Remove all rules from current profile
     Clear,
     #[command(hide = true)]

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

@@ -9,6 +9,10 @@ use crossterm::{
 };
 use dialoguer::Select;
 
+use crate::auth::builder_id::{
+    BuilderIdToken,
+    TokenType,
+};
 use crate::cli::chat::{
     ChatError,
     ChatSession,
@@ -98,10 +102,23 @@ impl ModelArgs {
     }
 }
 
-/// Currently, Sonnet 4 is set as the default model for non-FRA users.
-pub fn default_model_id(os: &Os) -> &'static str {
-    match os.database.get_auth_profile() {
-        Ok(Some(profile)) if profile.arn.split(':').nth(3) == Some("eu-central-1") => "CLAUDE_3_7_SONNET_20250219_V1_0",
-        _ => "CLAUDE_SONNET_4_20250514_V1_0",
+/// Returns Claude 3.7 for: Amazon IDC users, FRA region users
+/// Returns Claude 4.0 for: Builder ID users, other regions
+pub async fn default_model_id(os: &Os) -> &'static str {
+    // Check FRA region first
+    if let Ok(Some(profile)) = os.database.get_auth_profile() {
+        if profile.arn.split(':').nth(3) == Some("eu-central-1") {
+            return "CLAUDE_3_7_SONNET_20250219_V1_0";
+        }
+    }
+
+    // Check if Amazon IDC user
+    if let Ok(Some(token)) = BuilderIdToken::load(&os.database).await {
+        if matches!(token.token_type(), TokenType::IamIdentityCenter) && token.is_amzn_user() {
+            return "CLAUDE_3_7_SONNET_20250219_V1_0";
+        }
     }
+
+    // Default to 4.0
+    "CLAUDE_SONNET_4_20250514_V1_0"
 }

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

@@ -496,16 +496,10 @@ 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, os: &Os) -> Result<bool, ChatError> {
-        Ok(self
-            .backend_conversation_state(os, false, &mut vec![])
-            .await?
-            .history
-            .len()
-            >= 2)
+    pub async fn truncate_large_user_messages(&mut self) {
+        for (user_message, _) in &mut self.history {
+            user_message.truncate_safe(25_000);
+        }
     }
 
     /// Returns a [FigConversationState] capable of replacing the history of the current
@@ -559,14 +553,7 @@ impl ConversationState {
         };
 
         let conv_state = self.backend_conversation_state(os, false, &mut vec![]).await?;
-
-        // Include everything but the last message in the history.
-        let history_len = conv_state.history.len();
-        let history = if history_len < 2 {
-            vec![]
-        } else {
-            flatten_history(conv_state.history.take(history_len.saturating_sub(1)))
-        };
+        let history = flatten_history(conv_state.history);
 
         let user_input_message_context = UserInputMessageContext {
             env_state: Some(build_env_state()),

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

@@ -4,7 +4,10 @@ use serde::{
     Deserialize,
     Serialize,
 };
-use tracing::error;
+use tracing::{
+    error,
+    warn,
+};
 
 use super::consts::MAX_CURRENT_WORKING_DIRECTORY_LEN;
 use super::tools::{
@@ -15,6 +18,7 @@ use super::util::{
     document_to_serde_value,
     serde_value_to_document,
     truncate_safe,
+    truncate_safe_in_place,
 };
 use crate::api_client::model::{
     AssistantResponseMessage,
@@ -55,6 +59,30 @@ pub enum UserMessageContent {
     },
 }
 
+impl UserMessageContent {
+    fn truncate_safe(&mut self, max_bytes: usize) {
+        match self {
+            UserMessageContent::Prompt { prompt } => {
+                truncate_safe_in_place(prompt, max_bytes);
+            },
+            UserMessageContent::CancelledToolUses {
+                prompt,
+                tool_use_results,
+            } => {
+                if let Some(prompt) = prompt {
+                    truncate_safe_in_place(prompt, max_bytes / 2);
+                    truncate_safe_tool_use_results(tool_use_results.as_mut_slice(), max_bytes / 2);
+                } else {
+                    truncate_safe_tool_use_results(tool_use_results.as_mut_slice(), max_bytes);
+                }
+            },
+            UserMessageContent::ToolUseResults { tool_use_results } => {
+                truncate_safe_tool_use_results(tool_use_results.as_mut_slice(), max_bytes);
+            },
+        }
+    }
+}
+
 impl UserMessage {
     /// Creates a new [UserMessage::Prompt], automatically detecting and adding the user's
     /// environment [UserEnvContext].
@@ -193,6 +221,14 @@ impl UserMessage {
             UserMessageContent::ToolUseResults { .. } => None,
         }
     }
+
+    /// Truncates the content contained in this user message to a maximum length of `max_bytes`.
+    ///
+    /// This isn't a perfect truncation - JSON tool use results are ignored, and only the content
+    /// of the user message is truncated, ignoring extra context fields.
+    pub fn truncate_safe(&mut self, max_bytes: usize) {
+        self.content.truncate_safe(max_bytes);
+    }
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -225,6 +261,20 @@ impl From<ToolUseResult> for ToolResult {
     }
 }
 
+fn truncate_safe_tool_use_results(tool_use_results: &mut [ToolUseResult], max_bytes: usize) {
+    let max_bytes = max_bytes / tool_use_results.len();
+    for result in tool_use_results {
+        for content in &mut result.content {
+            match content {
+                ToolUseResultBlock::Json(_) => {
+                    warn!("Unable to truncate JSON safely");
+                },
+                ToolUseResultBlock::Text(t) => truncate_safe_in_place(t, max_bytes),
+            }
+        }
+    }
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub enum ToolUseResultBlock {
     Json(serde_json::Value),