You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Windows: the same archived conversation is indexed under two path spellings (C-- vs c--), showing as orphans and duplicates #202
What happens. On Windows, Claude Code's project folders can be named with the drive letter in either case: C--Users-… or c--Users-…. VS Code launches with c:\, the CLI and the desktop app with C:\. NTFS treats both names as the same folder, and the archive folder keeps whichever spelling was seen first. But exchanges.archive_path is stored with the spelling of the run that indexed it, and verifyIndex compares paths as exact strings (foundFiles.has(exchange.archivePath)).
Consequences, seen on one machine (v1.0.15 history, now on v1.6.0):
84 archived conversations (246 exchanges) recorded as …\c--Users-…\<id>.jsonl, while the folder on disk is C--Users-…. verify reports all of them as orphaned, and repair would delete them even though the files exist.
9 of those conversations were also indexed under the on-disk spelling, so their 159 exchanges come back twice in search results.
Suggested fix. On win32, normalize an archive path to its on-disk case before storing and before comparing it, for example with fs.realpathSync.native(p), or compare paths case-insensitively there. A one-off cleanup could then drop the rows whose archive_path differs from the on-disk spelling only by case, and let repair re-index what is left without rows (#200 / #201).
Activity