From e0a8738980965411f514f4a62c09f941efdea90c Mon Sep 17 00:00:00 2001
From: Steven Enamakel <enamakel@tinyhumans.ai>
Date: Tue, 28 Jul 2026 11:43:55 +0300
Subject: [PATCH] fix(memory): guard chunk tag rewrites

---
 src/memory/store/content/tags.rs       | 28 +++++++++++++++++++++++
 src/memory/store/content/tags_tests.rs | 31 ++++++++++++++++++++++++++
 2 files changed, 59 insertions(+)

diff --git a/src/memory/store/content/tags.rs b/src/memory/store/content/tags.rs
index 0a171dc..1f15ad9 100644
--- a/src/memory/store/content/tags.rs
+++ b/src/memory/store/content/tags.rs
@@ -35,6 +35,8 @@ pub fn update_chunk_tags(abs_path: &Path, tags: &[String]) -> anyhow::Result<()>
     let new_bytes = rewrite_tags(&old_bytes, &augmented)
         .map_err(|e| anyhow::anyhow!("rewrite_tags {:?}: {e}", abs_path))?;
 
+    ensure_tag_rewrite_preserves_body(&old_bytes, &new_bytes, abs_path)?;
+
     let parent = abs_path.parent().unwrap_or_else(|| Path::new("."));
     let tmp_name = format!(".tmp_tags_{}.md", crate_temp_id());
     let tmp_path = parent.join(&tmp_name);
@@ -57,6 +59,32 @@ pub fn update_chunk_tags(abs_path: &Path, tags: &[String]) -> anyhow::Result<()>
     Ok(())
 }
 
+/// Reject a front-matter rewrite unless both files parse and their bodies are
+/// byte-identical.
+///
+/// The body hash is persisted separately from the markdown file, so silently
+/// accepting an invalid or changed body would corrupt later retrieval.
+fn ensure_tag_rewrite_preserves_body(
+    old_bytes: &[u8],
+    new_bytes: &[u8],
+    abs_path: &Path,
+) -> anyhow::Result<()> {
+    let body = |bytes: &[u8]| -> Option<String> {
+        std::str::from_utf8(bytes)
+            .ok()
+            .and_then(split_front_matter)
+            .map(|(_, body)| body.to_owned())
+    };
+
+    match (body(old_bytes), body(new_bytes)) {
+        (Some(old), Some(new)) if old == new => Ok(()),
+        _ => Err(anyhow::anyhow!(
+            "tag rewrite would mutate or invalidate the body for {:?}",
+            abs_path
+        )),
+    }
+}
+
 /// Slugify an entity kind string for an Obsidian hierarchical tag.
 ///
 /// Output: lowercase, non-alphanumeric → `-`, collapsed, trimmed.
diff --git a/src/memory/store/content/tags_tests.rs b/src/memory/store/content/tags_tests.rs
index cd8bec1..1cac1e5 100644
--- a/src/memory/store/content/tags_tests.rs
+++ b/src/memory/store/content/tags_tests.rs
@@ -88,6 +88,37 @@ fn update_chunk_tags_is_noop_for_missing_file() {
     assert!(update_chunk_tags(&path, &["p/X".into()]).is_ok());
 }
 
+#[test]
+fn update_chunk_tags_rejects_unparseable_front_matter_without_overwriting() {
+    let dir = TempDir::new().unwrap();
+    let path = dir.path().join("malformed.md");
+    let original = b"not front matter\nBODY";
+    std::fs::write(&path, original).unwrap();
+
+    let result = update_chunk_tags(&path, &["person/Alice".into()]);
+
+    assert!(result.is_err());
+    assert_eq!(std::fs::read(&path).unwrap(), original);
+}
+
+#[test]
+fn body_guard_accepts_front_matter_only_changes() {
+    let path = Path::new("chunk.md");
+    let old = b"---\ntags: []\n---\nBODY";
+    let new = b"---\ntags:\n  - person/Alice\n---\nBODY";
+
+    assert!(ensure_tag_rewrite_preserves_body(old, new, path).is_ok());
+}
+
+#[test]
+fn body_guard_rejects_body_drift() {
+    let path = Path::new("chunk.md");
+    let old = b"---\ntags: []\n---\nBODY";
+    let new = b"---\ntags:\n  - person/Alice\n---\nDIFFERENT";
+
+    assert!(ensure_tag_rewrite_preserves_body(old, new, path).is_err());
+}
+
 #[test]
 fn slugify_tag_kind_examples() {
     assert_eq!(slugify_tag_kind("Person"), "person");