diff --git a/.jules/sentinel.md b/.jules/sentinel.md
new file mode 100644
index 000000000..50ae3b342
--- /dev/null
+++ b/.jules/sentinel.md
@@ -0,0 +1,4 @@
+## 2025-05-14 - Arithmetic Hardening in Entry Parsing and Splitting
+**Vulnerability:** Potential panics from integer overflows during archive entry parsing and multi-part archive splitting.
+**Learning:** Cumulative size calculations (like `compressed_size` or total entry length) and counters (like archive part numbers) in archive logic are vulnerable to overflows when processing large or malformed archives, especially on 32-bit platforms where `usize::MAX` is smaller.
+**Prevention:** Use `saturating_add` for size and length calculations to safely maintain upper bounds without crashing, and `checked_add` for structural fields like sequence numbers where an explicit error is preferable to wrap-around or saturation.
diff --git a/cli/src/command/core.rs b/cli/src/command/core.rs
index 28e0add9b..32e6937d8 100644
--- a/cli/src/command/core.rs
+++ b/cli/src/command/core.rs
@@ -1812,27 +1812,29 @@ where
         )
         .into());
     }
-    let mut part_num = 1;
+    let mut part_num: usize = 1;
     let mut writer = Archive::write_header(initial_writer)?;
 
     // NOTE: max_file_size - (PNA_HEADER + AHED + ANXT + AEND)
     let max_file_size = max_file_size - SPLIT_ARCHIVE_OVERHEAD_BYTES;
-    let mut written_entry_size = 0;
+    let mut written_entry_size: usize = 0;
     for entry in entries {
         let p = EntryPart::from(entry?);
         let parts = split_to_parts(
             p.as_ref(),
-            max_file_size - written_entry_size,
+            max_file_size.saturating_sub(written_entry_size),
             max_file_size,
         )?;
         for part in parts {
-            if written_entry_size + part.bytes_len() > max_file_size {
-                part_num += 1;
+            if written_entry_size.saturating_add(part.bytes_len()) > max_file_size {
+                part_num = part_num.checked_add(1).ok_or_else(|| {
+                    io::Error::new(io::ErrorKind::InvalidData, "archive part number overflow")
+                })?;
                 let file = get_next_writer(part_num)?;
                 writer = writer.split_to_next_archive(file)?;
                 written_entry_size = 0;
             }
-            written_entry_size += writer.add_entry_part(part)?;
+            written_entry_size = written_entry_size.saturating_add(writer.add_entry_part(part)?);
         }
     }
     writer.finalize()?;
diff --git a/lib/src/entry.rs b/lib/src/entry.rs
index e36f24aa3..c0e261fb7 100644
--- a/lib/src/entry.rs
+++ b/lib/src/entry.rs
@@ -633,7 +633,7 @@ where
                 ),
             ));
         }
-        let mut compressed_size = 0;
+        let mut compressed_size: usize = 0;
         let mut extra = vec![];
         let mut data = vec![];
         let mut xattrs = vec![];
@@ -657,7 +657,7 @@ where
                     );
                 }
                 ChunkType::FDAT => {
-                    compressed_size += chunk.data().len();
+                    compressed_size = compressed_size.saturating_add(chunk.data().len());
                     data.push(chunk.data);
                 }
                 ChunkType::fSIZ => size = Some(u128_from_be_bytes_last(chunk.data())),
@@ -1132,7 +1132,9 @@ where
     /// Returns the total length of this entry part in bytes.
     #[inline]
     pub fn bytes_len(&self) -> usize {
-        self.0.iter().map(|chunk| chunk.bytes_len()).sum()
+        self.0
+            .iter()
+            .fold(0, |acc, chunk| acc.saturating_add(chunk.bytes_len()))
     }
 
     /// Get reference.
@@ -1156,11 +1158,18 @@ impl EntryPart<&[u8]> {
         }
         let mut remaining = VecDeque::from(self.0);
         let mut first = Vec::new();
-        let mut total_size = 0;
+        let mut total_size: usize = 0;
         while let Some(chunk) = remaining.pop_front() {
             // NOTE: If over max size, restore to the remaining chunk
-            if max_bytes_len < total_size + chunk.bytes_len() {
-                if chunk.is_stream_chunk() && total_size + MIN_CHUNK_BYTES_SIZE < max_bytes_len {
+            if total_size
+                .checked_add(chunk.bytes_len())
+                .is_none_or(|sum| max_bytes_len < sum)
+            {
+                if chunk.is_stream_chunk()
+                    && total_size
+                        .checked_add(MIN_CHUNK_BYTES_SIZE)
+                        .is_some_and(|sum| sum < max_bytes_len)
+                {
                     let available_bytes_len = max_bytes_len - total_size;
                     let chunk_split_index = available_bytes_len - MIN_CHUNK_BYTES_SIZE;
                     let (x, y) = chunk_data_split(chunk.ty, chunk.data, chunk_split_index);
@@ -1173,7 +1182,7 @@ impl EntryPart<&[u8]> {
                 }
                 break;
             }
-            total_size += chunk.bytes_len();
+            total_size = total_size.saturating_add(chunk.bytes_len());
             first.push(chunk);
         }
         if first.is_empty() {