Relationships
#4753 swamp start: finds more Linux terminals (Ptyxis and others), a failed launch says what it tried, and resume reopens the conversation start launched
Opened by skunk-ape · 11/14/2026· Shipped 11/14/2026
Found by an external tester running swamp start on Fedora Atomic (Bluefin or Silverblue style: home under /var/home, Homebrew under /home/linuxbrew) with Codex. Affects every swamp start user, not only Swampify. #4751 (herdr, T3 Code and a copied prompt as launch targets) builds on this.
Repo: swamp (swamp start and its terminal launcher only).
What happened
- On the resume screen, start couldn't open a terminal and showed the copy-command steps.
- Pressing Launch agent made the page flicker and come back unchanged. Nothing appeared on the page or in the terminal running
swamp start. - The copied command resumes with
codex resume --last, which takes the newest Codex session. The tester also runs Codex through other apps, so that may not be the conversation start launched.
Causes
- Terminal list. On Linux,
SystemTerminalLauncher(src/infrastructure/process/terminal/terminal_launcher.ts) tries$TERMINAL,x-terminal-emulator,gnome-terminal,konsoleandxfce4-terminal. Fedora Atomic desktops default to Ptyxis, which isn't on the list, and nor are several common terminals. - Silent retry. Launch agent tries the same list again. A second
copy_commandoutcome renders the same screen, and nothing is logged. - Resume target. Start records no conversation id.
claude --continueandcodex resume --lasttake the newest conversation, which can belong to another app, a Swampify reading run, or another session in the folder.
Changes
1. More Linux terminals
Add, after $TERMINAL and before the current fallbacks, each with the flag it needs to run the launcher script (verify each one):
ptyxis(Fedora 41+ and Atomic desktops)kgx(GNOME Console)kitty,alacritty,ghostty,wezterm(wezterm start --),foot
If swamp start runs inside a toolbox or distrobox container, the host's terminals aren't on its PATH. Detect that (/run/.containerenv or /run/.toolboxenv) and say so in the failure notice rather than failing silently. Opening a host terminal from inside the container is out of scope.
2. A failed launch says what it tried
- When no terminal opens, the outcome lists the terminals it tried and why each failed (not found, exited with an error).
- The page shows that in one line under the copy-command steps, for example "Couldn't open a terminal. Tried: x-terminal-emulator (not found), gnome-terminal (not found). Set
TERMINALto your terminal, or copy the command." A retry that fails again says so, rather than re-rendering the same screen. - The terminal running
swamp startlogs the same at warning level.
3. Resume the conversation start launched
- Claude Code. Launch a task with
--session-id <uuid>that start generates, and keep it on the launch record. Resume with--resume <uuid>. Fall back to--continuewhen the record has none (older launches). - Codex. At resume time, find the newest Codex session file whose recorded working directory is this workspace and which started at or after start's launch, and run
codex resume <id>. Read only the session's header line. Fall back toresume --lastwhen none is found. - Other agents keep their current resume behaviour.
- The launch record gains an optional session id. Older records without one stay readable.
Out of scope
- herdr, T3 Code and the copied prompt as launch targets (#4751).
- Opening a host terminal from inside a container.
Done when
- Unit tests cover the terminal order with each new terminal present or absent and
$TERMINALset, the args each one gets, and the container notice. - Unit tests cover the failure outcome listing what was tried, its rendering on the page, a repeated failure that says so, and the warning log.
- Unit tests cover Claude's session id on launch and resume, the Codex lookup (matching folder, a newer session elsewhere ignored, none found falls back to
--last), and older launch records. - An integration test through the start server launches and resumes with a fake Claude and a fake Codex, and checks the resume command names the launched session.
- A manual check on a Fedora Atomic desktop with Ptyxis (or a VM): Launch agent opens Ptyxis, and resume reopens the launched conversation after another Codex session ran elsewhere.
Shipped
Click a lifecycle step above to view its details.
Sign in to post a ripple.