termdeck
Claude Code fix: A session you know exists is not in the list
Sessions are grouped by a slug built from the working directory string, character by character. Two paths that open the same folder are therefore two different projects.
Comes from Claude Code.
Diagnosis
Each character that is not a letter or a digit becomes a hyphen, so the directory name is a transcription of the path rather than a lookup. Different text means a different folder, even when both paths reach the same files.
Starting the CLI through a symlinked route, or inside a git worktree, produces a different working directory string and therefore a different project folder. The sessions are not lost, they are filed somewhere you have not looked.
A path written with and without its final separator encodes two ways, which is why a careful reader normalises before slugging and a careless one ends up with a pair of near-identical folders.
A hyphen in the folder name might have been a hyphen in the path or a separator between parts. That is why the real path belongs with the session data, and why decoding the folder name is a last resort rather than a lookup.
Check
| Run this | What a good answer looks like |
|---|---|
pwd | The exact string that decides the folder. Run it where the session was started, not where you happen to be now. |
ls ~/.claude/projects | One directory per project path the CLI has seen. The one you want is usually a near twin of the one you expected, differing by a symlink component or a case. |
ls -t ~/.claude/projects/*/*.jsonl | head | Sorted by time, so the session you just ran is at the top whichever folder it landed in. |
Every command here has been run on a real machine. Termdeck answers the same questions for every machine you have connected, without opening a terminal on any of them.
Fix
Resolve the symlinks once, use that path from then on, and the sessions rejoin. On a machine whose home is reachable two ways, picking one and staying with it is the entire fix.
The transcript is a plain JSONL file and its name carries the session id, which is what any resume takes.
Termdeck groups sessions by the working directory recorded inside the transcript rather than by decoding a folder name, which is what keeps a worktree and its parent legible side by side.
Not this
The CLI writes to whichever folder its slug rule produces. A hand-renamed directory is ignored on the next run, and now you have two.
FAQ
No. They stay as plain JSONL files until something removes them.
They are different working directories, so they slug differently. The grouping is per path rather than per repository.
The file is portable and its id travels in the name. Nothing rewrites the path recorded inside it, so a moved session still remembers where it ran.
Claude Code, Codex and Grok in one browser console