Only the latest published version of windowkeeper receives security fixes.
Please do NOT open a public issue for security problems. Instead use GitHub private vulnerability reporting or email yoad.elkayam@gmail.com.
You can expect an acknowledgment within a few days. Please include a reproduction if you can.
windowkeeper sits between an agent harness and the model API, so the interesting surface is:
- The archive stores every message and tool result verbatim, by design. Treat any persistent archive (file, SQLite) as sensitive: it contains everything the model ever saw, including data later evicted from the window.
- The ledger records identifiers extracted from user messages (file paths, symbol names, error strings) to explain re-surfacing decisions. Treat ledger files as sensitive.
- Re-surfacing by identifier match reads from the archive only. It never executes tools. Serving a repeated tool call from the archive is opt-in per tool and never applies to tools listed as volatile.