checkpoint

will need this for it to work

.cargo/config.toml

@@ -0,0 +1,5 @@
+[target.aarch64-apple-darwin]
+rustflags = ["-C", "link-args=-Wl,-rpath,@executable_path"]
+
+[target.x86_64-apple-darwin]
+rustflags = ["-C", "link-args=-Wl,-rpath,@executable_path"]

README.md

@@ -136,8 +136,12 @@ G3 is designed for:
 # Build the project
 cargo build --release
 
-# Run G3
-cargo run
+# Run from the build directory
+./target/release/g3
+
+# Or copy both files to somewhere in your PATH (macOS only needs both files)
+cp target/release/g3 ~/.local/bin/
+cp target/release/libVisionBridge.dylib ~/.local/bin/  # macOS only
 
 # Execute a task
 g3 "implement a function to calculate fibonacci numbers"

build.rs

@@ -1,11 +0,0 @@
-use std::env;
-
-fn main() {
-    // Only add rpaths on macOS
-    if env::var("CARGO_CFG_TARGET_OS").unwrap() == "macos" {
-        // Add rpath so libVisionBridge.dylib can be found at runtime
-        // @executable_path means "relative to the executable"
-        println!("cargo:rustc-link-arg=-Wl,-rpath,@executable_path");
-        println!("cargo:rustc-link-arg=-Wl,-rpath,@loader_path");
-    }
-}

crates/g3-cli/build.rs

@@ -1,11 +0,0 @@
-use std::env;
-
-fn main() {
-    // Only add rpaths on macOS
-    if env::var("CARGO_CFG_TARGET_OS").unwrap() == "macos" {
-        // Add rpath so libVisionBridge.dylib can be found at runtime
-        // @executable_path means "relative to the executable"
-        println!("cargo:rustc-link-arg=-Wl,-rpath,@executable_path");
-        println!("cargo:rustc-link-arg=-Wl,-rpath,@loader_path");
-    }
-}

crates/g3-core/src/lib.rs

@@ -1444,6 +1444,13 @@ Template:
                 let available = model_limit
                     .saturating_sub(current_usage)
                     .saturating_sub(5000);
+                // Ensure we have at least 1 token available, otherwise we can't generate a summary
+                if available == 0 {
+                    return Err(anyhow::anyhow!(
+                        "Insufficient tokens available for summary generation. Current usage: {} tokens, model limit: {} tokens. Please start a new session.",
+                        current_usage, model_limit
+                    ));
+                }
                 Some(available.min(10_000))
             }
             "embedded" => {
@@ -1452,10 +1459,24 @@ Template:
                 let available = model_limit
                     .saturating_sub(current_usage)
                     .saturating_sub(1000);
+                // Ensure we have at least 1 token available
+                if available == 0 {
+                    return Err(anyhow::anyhow!(
+                        "Insufficient tokens available for summary generation. Current usage: {} tokens, model limit: {} tokens. Please start a new session.",
+                        current_usage, model_limit
+                    ));
+                }
                 Some(available.min(3000))
             }
             _ => {
                 let available = self.context_window.remaining_tokens().saturating_sub(2000);
+                // Ensure we have at least 1 token available
+                if available == 0 {
+                    return Err(anyhow::anyhow!(
+                        "Insufficient tokens available for summary generation. Current usage: {} tokens. Please start a new session.",
+                        self.context_window.used_tokens
+                    ));
+                }
                 Some(available.min(5000))
             }
         };
@@ -2347,6 +2368,13 @@ Template:
                     let available = model_limit
                         .saturating_sub(current_usage)
                         .saturating_sub(5000);
+                    // Ensure we have at least 1 token available, otherwise we can't generate a summary
+                    if available == 0 {
+                        return Err(anyhow::anyhow!(
+                            "Insufficient tokens available for summary generation. Current usage: {} tokens, model limit: {} tokens. Please start a new session.",
+                            current_usage, model_limit
+                        ));
+                    }
                     // Cap at a reasonable summary size (10k tokens max)
                     Some(available.min(10_000))
                 }
@@ -2358,12 +2386,26 @@ Template:
                     let available = model_limit
                         .saturating_sub(current_usage)
                         .saturating_sub(1000);
+                    // Ensure we have at least 1 token available
+                    if available == 0 {
+                        return Err(anyhow::anyhow!(
+                            "Insufficient tokens available for summary generation. Current usage: {} tokens, model limit: {} tokens. Please start a new session.",
+                            current_usage, model_limit
+                        ));
+                    }
                     // Cap at 3k for embedded models
                     Some(available.min(3000))
                 }
                 _ => {
                     // Default: conservative approach
                     let available = self.context_window.remaining_tokens().saturating_sub(2000);
+                    // Ensure we have at least 1 token available
+                    if available == 0 {
+                        return Err(anyhow::anyhow!(
+                            "Insufficient tokens available for summary generation. Current usage: {} tokens. Please start a new session.",
+                            self.context_window.used_tokens
+                        ));
+                    }
                     Some(available.min(5000))
                 }
             };
@@ -2675,12 +2717,7 @@ Template:
 
                             // Display tool execution result with proper indentation
                             if tool_call.tool != "final_output" {
-                                // Skip displaying output for shell tool since it was already streamed
-                                let should_display_output = tool_call.tool != "shell";
-                                
-                                let output_lines: Vec<&str> = if should_display_output {
-                                    tool_result.lines().collect()
-                                } else { vec![] };
+                                let output_lines: Vec<&str> = tool_result.lines().collect();
 
                                 // Check if UI wants full output (machine mode) or truncated (human mode)
                                 let wants_full = self.ui_writer.wants_full_output();
@@ -3192,16 +3229,13 @@ Template:
                         {
                             Ok(result) => {
                                 if result.success {
-                                    // Don't return stdout - it was already streamed to the UI
-                                    // Returning it would cause duplicate output
-                                    Ok("✅ Command executed successfully".to_string())
+                                    Ok(if result.stdout.is_empty() {
+                                        "✅ Command executed successfully".to_string()
                                     } else {
-                                    // For errors, return stderr since it wasn't streamed
-                                    Ok(if result.stderr.is_empty() {
-                                        "❌ Command failed".to_string()
-                                    } else {
-                                        format!("❌ Command failed: {}", result.stderr.trim())
+                                        result.stdout.trim().to_string()
                                     })
+                                } else {
+                                    Ok(format!("❌ Command failed: {}", result.stderr.trim()))
                                 }
                             }
                             Err(e) => Ok(format!("❌ Execution error: {}", e)),