diff --git a/src/main.rs b/src/main.rs
index 93dffbe6c..4caf60e32 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -579,8 +579,14 @@ fn cmd_auth(config_path: Option<std::path::PathBuf>, auth_cmd: AuthCommand) -> a
                         } else {
                             eprintln!("Anthropic OAuth: valid (expires in {}m)", expires_min);
                         }
-                        eprintln!("  access token: {}...", &creds.access_token[..20]);
-                        eprintln!("  refresh token: {}...", &creds.refresh_token[..20]);
+                        eprintln!(
+                            "  access token: <redacted> ({} bytes)",
+                            creds.access_token.len()
+                        );
+                        eprintln!(
+                            "  refresh token: <redacted> ({} bytes)",
+                            creds.refresh_token.len()
+                        );
                         eprintln!(
                             "  credentials file: {}",
                             spacebot::auth::credentials_path(&instance_dir).display()
diff --git a/src/tools.rs b/src/tools.rs
index f1e53257b..677bd3e85 100644
--- a/src/tools.rs
+++ b/src/tools.rs
@@ -992,4 +992,87 @@ mod tests {
         assert_eq!(truncate_utf8_ellipsis(text, 10), "🙂...");
         assert!(truncate_output(text, 5).starts_with("🙂"));
     }
+
+    #[test]
+    fn truncate_cyrillic_does_not_panic() {
+        // Cyrillic chars are 2 bytes each in UTF-8
+        let text = "Привет, мир!"; // "Hello, world!" in Russian
+        // Cutting at byte 5 would land inside 'и' (bytes 4..5) - this must not panic
+        let result = truncate_utf8_ellipsis(text, 5);
+        assert!(!result.is_empty());
+        // "П" (2 bytes) + "..." (3 bytes) = 5, fits exactly
+        assert_eq!(result, "П...");
+
+        let result = truncate_utf8_ellipsis(text, 10);
+        assert!(!result.is_empty());
+        // Should truncate to a valid char boundary and append "..."
+        assert!(result.ends_with("..."));
+    }
+
+    #[test]
+    fn truncate_cjk_does_not_panic() {
+        // CJK chars are 3 bytes each in UTF-8
+        let text = "你好世界测试"; // "Hello world test" in Chinese
+        // Cutting at byte 4 would land inside '好' (bytes 3..5)
+        let result = truncate_utf8_ellipsis(text, 4);
+        assert_eq!(result, "你"); // only 3 bytes fit, no room for "..."
+
+        let result = truncate_utf8_ellipsis(text, 10);
+        assert!(!result.is_empty());
+        // 2 CJK chars (6 bytes) + "..." (3 bytes) = 9 bytes fits in 10
+        assert_eq!(result, "你好...");
+    }
+
+    #[test]
+    fn truncate_emoji_does_not_panic() {
+        // Emoji are 4 bytes each in UTF-8
+        let text = "Hello 😀🎉🚀 World";
+        let result = truncate_utf8_ellipsis(text, 10);
+        assert!(!result.is_empty());
+        // "Hello " (6 bytes) + "😀" won't fit with "..." in 10 bytes
+        // "Hello " (6 bytes) + "..." (3 bytes) = 9 bytes fits
+        assert_eq!(result, "Hello ...");
+
+        // Larger budget
+        let result = truncate_utf8_ellipsis(text, 15);
+        assert!(!result.is_empty());
+        assert!(result.ends_with("..."));
+    }
+
+    #[test]
+    fn truncate_mixed_multibyte_content() {
+        // Mix of ASCII, Cyrillic (2-byte), CJK (3-byte), and emoji (4-byte)
+        let text = "Hi Привет 你好 😀";
+        let result = truncate_utf8_ellipsis(text, 20);
+        assert!(!result.is_empty());
+        assert!(result.ends_with("..."));
+        // Verify it's valid UTF-8 by iterating chars
+        assert!(result.chars().count() > 0);
+    }
+
+    #[test]
+    fn truncate_output_multibyte_does_not_panic() {
+        let cyrillic = "Привет, мир! Это тестовая строка для проверки.";
+        let result = truncate_output(cyrillic, 15);
+        assert!(!result.is_empty());
+
+        let cjk = "你好世界，这是一个测试字符串。";
+        let result = truncate_output(cjk, 10);
+        assert!(!result.is_empty());
+
+        let emoji = "🎉🚀😀🌍💻🔥";
+        let result = truncate_output(emoji, 6);
+        assert!(!result.is_empty());
+    }
+
+    #[test]
+    fn truncate_at_exact_char_boundary_works() {
+        let text = "абв"; // 3 Cyrillic chars, 6 bytes total
+        // Exactly at a char boundary (4 bytes = 2 chars)
+        let result = truncate_utf8_ellipsis(text, 6);
+        assert_eq!(result, "абв"); // fits entirely
+
+        let result = truncate_utf8_ellipsis(text, 7);
+        assert_eq!(result, "абв"); // also fits, no truncation needed
+    }
 }
diff --git a/src/tools/browser.rs b/src/tools/browser.rs
index d0590846c..f5affe75f 100644
--- a/src/tools/browser.rs
+++ b/src/tools/browser.rs
@@ -529,11 +529,7 @@ fn render_snapshot_node(node: &SnapshotNode, depth: usize, output: &mut String)
     if !node.name.is_empty() {
         output.push_str(" \"");
         // Truncate very long names for context efficiency.
-        let display_name = if node.name.len() > 200 {
-            format!("{}...", &node.name[..200])
-        } else {
-            node.name.clone()
-        };
+        let display_name = super::truncate_utf8_ellipsis(&node.name, 200);
         output.push_str(&display_name.replace('"', "\\\""));
         output.push('"');
     }
@@ -574,11 +570,7 @@ fn render_snapshot_node(node: &SnapshotNode, depth: usize, output: &mut String)
 
     // Value (e.g., text input current value)
     if let Some(ref value) = node.value {
-        let display_value = if value.len() > 100 {
-            format!("{}...", &value[..100])
-        } else {
-            value.clone()
-        };
+        let display_value = super::truncate_utf8_ellipsis(value, 100);
         output.push_str(&format!(
             " value=\"{}\"",
             display_value.replace('"', "\\\"")
@@ -1588,11 +1580,7 @@ impl Tool for BrowserTypeTool {
                 args.secret.as_deref().unwrap_or("unknown")
             )
         } else {
-            let display_text = if text_value.len() > 50 {
-                format!("{}...", &text_value[..50])
-            } else {
-                text_value
-            };
+            let display_text = super::truncate_utf8_ellipsis(&text_value, 50);
             format!("Typed '{display_text}' into element at {label}")
         };