From cb0e225e7864108350786737a8c703b8a2cc0425 Mon Sep 17 00:00:00 2001
From: Hermes Agent <linux2010@users.noreply.github.com>
Date: Tue, 21 Apr 2026 08:43:51 +0000
Subject: [PATCH] fix(skills_hub): handle bytes content in bundle_content_hash

What broke:
Running `hermes skills update` crashed with AttributeError: 'bytes'
object has no attribute 'encode' when skill bundles contained binary
files (e.g. audio assets from official skills).

Root cause:
SkillBundle.files is defined as Dict[str, Union[str, bytes]], but
bundle_content_hash() assumed all file content was str and called
.encode("utf-8") on every value. When the value was already bytes,
this caused a crash.

Why this fix is minimal:
Single function fix - detect type and route bytes directly to
hash.update(), encode strings as before. No broader changes to
SkillBundle, file handling, or any other hash computation paths.

What I tested:
Added regression test verifying bytes content is handled correctly
and produces deterministic hashes.

What I intentionally did not change:
- No changes to SkillBundle class or file loading logic
- No changes to other hash computation functions
- No changes to quarantine/installation logic
---
 tests/tools/test_skills_hub.py | 29 +++++++++++++++++++++++++++++
 tools/skills_hub.py            |  7 ++++++-
 2 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/tests/tools/test_skills_hub.py b/tests/tools/test_skills_hub.py
index 24d1e87affcc8..32115b9a254c7 100644
--- a/tests/tools/test_skills_hub.py
+++ b/tests/tools/test_skills_hub.py
@@ -674,6 +674,35 @@ def fake_get(url, *args, **kwargs):
 
 
 class TestCheckForSkillUpdates:
+    def test_bundle_content_hash_handles_bytes_content(self):
+        """Regression test for #2739: bundle_content_hash must handle bytes files."""
+        # SkillBundle.files can contain Union[str, bytes]
+        bundle = SkillBundle(
+            name="binary-skill",
+            files={
+                "SKILL.md": "---\nname: test\n---\n",
+                "assets/audio.wav": b"RIFF\x00\x01fakewav",  # bytes content
+            },
+            source="github",
+            identifier="owner/repo/binary-skill",
+            trust_level="community",
+        )
+        # Should not crash with AttributeError: 'bytes' object has no attribute 'encode'
+        hash_result = bundle_content_hash(bundle)
+        assert hash_result.startswith("sha256:")
+        # Deterministic: same content = same hash
+        bundle2 = SkillBundle(
+            name="binary-skill",
+            files={
+                "SKILL.md": "---\nname: test\n---\n",
+                "assets/audio.wav": b"RIFF\x00\x01fakewav",
+            },
+            source="github",
+            identifier="owner/repo/binary-skill",
+            trust_level="community",
+        )
+        assert bundle_content_hash(bundle2) == hash_result
+
     def test_bundle_content_hash_matches_installed_content_hash(self, tmp_path):
         from tools.skills_guard import content_hash
 
diff --git a/tools/skills_hub.py b/tools/skills_hub.py
index 47aef8075b7c1..8724ff7aeadc2 100644
--- a/tools/skills_hub.py
+++ b/tools/skills_hub.py
@@ -2630,7 +2630,12 @@ def bundle_content_hash(bundle: SkillBundle) -> str:
     """Compute a deterministic hash for an in-memory skill bundle."""
     h = hashlib.sha256()
     for rel_path in sorted(bundle.files):
-        h.update(bundle.files[rel_path].encode("utf-8"))
+        content = bundle.files[rel_path]
+        # files can contain str or bytes; handle both
+        if isinstance(content, bytes):
+            h.update(content)
+        else:
+            h.update(content.encode("utf-8"))
     return f"sha256:{h.hexdigest()[:16]}"