What happened?

tool_add_drawer generates the drawer ID by hashing only the first 100 characters of the content (content[:100]). Two documents in the same wing and room whose first 100 characters are identical are assigned the same drawer ID. When the second document is filed, it silently overwrites the first via upsert — no error, no warning, no indication that data was lost.

What did you expect?

Each document with distinct content should receive a unique drawer ID regardless of how similar the opening characters are. Filing a second document should never overwrite a different existing document.

How to reproduce:

1. File a drawer with content that starts with a header longer than 100 characters (e.g. a company wiki template, Notion export, or meeting notes template)
2. File a second drawer in the same wing and room whose content shares the same first 100 characters but differs after that
3. Retrieve by the first drawer's ID — the original content is gone, replaced by the second document

import hashlib

wing, room = "work", "decisions"
header = "# ACME Corp Internal Knowledge Base\n**Project:** Alpha | **Team:** Backend | **Status:** Active\n\n"
doc1 = header + "Decision: Use PostgreSQL for primary storage."
doc2 = header + "Decision: Use Redis for session caching."

id1 = f"drawer_{wing}_{room}_{hashlib.sha256((wing + room + doc1[:100]).encode()).hexdigest()[:24]}"
id2 = f"drawer_{wing}_{room}_{hashlib.sha256((wing + room + doc2[:100]).encode()).hexdigest()[:24]}"
print(id1 == id2)  # True — data loss

Environment:

• OS: macOS 15
• Python version: 3.11.9
• MemPalace version: 3.1.0 (commit 068dbd9)