macos-plugin/skills/kitty-session-persistence/SKILL.md
Snapshot and restore kitty terminal sessions on macOS. Use when surviving WindowServer hangs without losing tabs, configuring snapshots, or restoring after a force-reboot.
npx skillsauth add laurigates/claude-plugins kitty-session-persistenceInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
| Use this skill when... | Use something else when... |
|------------------------|----------------------------|
| Terminal sessions need to survive WindowServer hangs and force-reboots | The user's terminal is iTerm2 / Terminal.app / Ghostty / Wezterm — kitty-specific |
| Configuring kitty's listen_on socket and a snapshot LaunchAgent | Cross-platform terminal config — this is macOS LaunchAgents |
| Restoring a previous session after a crash | One-off "open these tabs" — kitty @ launch directly is enough |
| Auditing whether snapshot capture is still running | General system diagnostics — see macos-incident-postmortem |
This skill is macOS-only. All commands assume Darwin and launchctl-managed LaunchAgents. Refuse to act if uname -s is not Darwin.
test "$(uname -s)" = "Darwin" || { echo "macos-plugin: not Darwin, refusing"; exit 1; }
Kitty exposes a remote control protocol via a Unix socket. With listen_on configured, kitten @ ls returns a JSON description of every OS window, tab, and window inside kitty — enough to reconstruct the layout from scratch.
The pattern this skill installs:
listen_on unix:/tmp/kitty-$USER in kitty.conf so the snapshot script can always reach the running instance.kitten @ ls output to a timestamped JSON file under ~/.local/state/kitty-sessions/, with a latest.json symlink.latest.json and re-launches tabs with the recorded cwd and command.allow_remote_control yes in kitty.conf or --allow-remote-control yes at launch.listen_on and the kitten @ --to flag the snapshot script uses.StandardOutPath and StandardErrorPath are essential — failing snapshots are otherwise silent.RunAtLoad: false and rely on StartInterval so a kitty restart doesn't spawn a backlog of immediate snapshots.Add to ~/.config/kitty/kitty.conf:
allow_remote_control yes
listen_on unix:/tmp/kitty-{kitty_pid}
Using {kitty_pid} makes each kitty instance addressable on its own socket. For the persistent-socket pattern this skill needs, use the user-stable form instead:
allow_remote_control yes
listen_on unix:/tmp/kitty-USER
Replace USER with the actual username. The snapshot LaunchAgent points at this exact path.
Reload kitty after editing:
kitty @ --to unix:/tmp/kitty-USER load-config
Create ~/.local/bin/kitty-snapshot (mode 0755):
#!/usr/bin/env bash
set -uo pipefail
[ "$(uname -s)" = "Darwin" ] || exit 0
SOCKET="unix:/tmp/kitty-${USER}"
STATE_DIR="${HOME}/.local/state/kitty-sessions"
mkdir -p "$STATE_DIR"
ts="$(date -u +%Y%m%dT%H%M%SZ)"
out="${STATE_DIR}/${ts}.json"
if ! kitten @ --to "$SOCKET" ls >"$out" 2>>"${STATE_DIR}/snapshot.err"; then
rm -f "$out"
exit 0 # kitty not running — silent success
fi
ln -sfn "$out" "${STATE_DIR}/latest.json"
# Keep ~7 days of 5-min snapshots = 2016 files; cap at 4032 (14 days)
ls -1t "$STATE_DIR"/*.json 2>/dev/null | tail -n +4033 | xargs -r rm -f
The exit 0 on kitten @ ls failure is intentional — the LaunchAgent fires every 5 minutes regardless of whether kitty is running.
Write ~/Library/LaunchAgents/com.<reverse-dns>.kitty-snapshot.plist:
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
"http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>Label</key>
<string>com.example.kitty-snapshot</string>
<key>ProgramArguments</key>
<array>
<string>/Users/USER/.local/bin/kitty-snapshot</string>
</array>
<key>StartInterval</key>
<integer>300</integer>
<key>RunAtLoad</key>
<false/>
<key>StandardOutPath</key>
<string>/Users/USER/.local/state/kitty-sessions/launchd.out</string>
<key>StandardErrorPath</key>
<string>/Users/USER/.local/state/kitty-sessions/launchd.err</string>
</dict>
</plist>
Replace com.example with a stable reverse-DNS prefix and USER with the literal username (LaunchAgents do not expand $HOME).
Validate, load, and verify:
plutil -lint ~/Library/LaunchAgents/com.example.kitty-snapshot.plist
launchctl bootstrap "gui/$(id -u)" ~/Library/LaunchAgents/com.example.kitty-snapshot.plist
launchctl print "gui/$(id -u)/com.example.kitty-snapshot" | grep -E 'state|last exit'
bootstrap replaces the deprecated launchctl load; on macOS 11+ prefer it.
After ~5 minutes:
ls -lt ~/.local/state/kitty-sessions/*.json | head -3
There should be JSON files newer than the LaunchAgent install time. The newest file's content should match what kitty has open right now.
Read latest.json and re-launch each window/tab. A minimal restore script:
#!/usr/bin/env bash
set -euo pipefail
[ "$(uname -s)" = "Darwin" ] || exit 1
LATEST="${HOME}/.local/state/kitty-sessions/latest.json"
SOCKET="unix:/tmp/kitty-${USER}"
[ -L "$LATEST" ] || { echo "no snapshot found"; exit 1; }
jq -c '.[] | .tabs[] | .windows[] | {cwd, foreground_processes}' "$LATEST" |
while IFS= read -r win; do
cwd=$(printf '%s' "$win" | jq -r '.cwd')
cmd=$(printf '%s' "$win" | jq -r '.foreground_processes[0].cmdline | join(" ")')
kitten @ --to "$SOCKET" launch --type=tab --cwd="$cwd" "$cmd"
done
The shape of kitten @ ls output is [{tabs:[{windows:[{cwd, foreground_processes:[{cmdline}]}]}]}]. Adjust the jq query if a future kitty release changes it.
ls -1t ~/.local/state/kitty-sessions/*.json | head -10
jq '. | length, [.[].tabs | length] | add' ~/.local/state/kitty-sessions/latest.json
# → window count, tab count
launchctl print "gui/$(id -u)/com.example.kitty-snapshot" | \
grep -E 'state|last exit|on demand'
state = running (briefly, every 5 minutes) and last exit code = 0 indicate health. Persistent non-zero exit means something is broken — check launchd.err.
launchctl bootout "gui/$(id -u)/com.example.kitty-snapshot"
bootout is the modern equivalent of launchctl unload.
| Interval | When to use |
|----------|-------------|
| 60s (StartInterval=60) | Heavy editing days — minimum loss-on-crash |
| 300s (default) | Balanced — typical workflow |
| 900s (15 min) | Stable, low-edit days — less noise in ls -lt |
Reload after editing the plist:
launchctl bootout "gui/$(id -u)/com.example.kitty-snapshot"
launchctl bootstrap "gui/$(id -u)" ~/Library/LaunchAgents/com.example.kitty-snapshot.plist
| Context | Command |
|---------|---------|
| Snapshot count last hour | find ~/.local/state/kitty-sessions -name '*.json' -newermt '1 hour ago' \| wc -l |
| Most recent snapshot age | stat -f '%Sm %N' ~/.local/state/kitty-sessions/latest.json |
| Validate a plist | plutil -lint ~/Library/LaunchAgents/com.example.kitty-snapshot.plist |
| Last exit code | launchctl print "gui/$(id -u)/com.example.kitty-snapshot" \| awk '/last exit code/{print $NF}' |
| Live socket reachable | kitten @ --to unix:/tmp/kitty-$USER ls \| jq '. \| length' |
| Operation | Command |
|-----------|---------|
| Load (modern) | launchctl bootstrap "gui/$(id -u)" <plist> |
| Unload (modern) | launchctl bootout "gui/$(id -u)/<label>" |
| Inspect | launchctl print "gui/$(id -u)/<label>" |
| Run now | launchctl kickstart -k "gui/$(id -u)/<label>" |
| List user agents | launchctl print-disabled "gui/$(id -u)" |
| Operation | Command |
|-----------|---------|
| List sessions | kitten @ --to <socket> ls |
| Launch tab | kitten @ --to <socket> launch --type=tab --cwd=<path> <cmd> |
| Reload config | kitten @ --to <socket> load-config |
| Send text | kitten @ --to <socket> send-text --match "title:foo" "ls\n" |
~/.local/state/kitty-sessions/
├── 20260503T120000Z.json
├── 20260503T120500Z.json
├── ...
├── latest.json -> 20260503T143000Z.json
├── snapshot.err
├── launchd.out
└── launchd.err
| Symptom | Cause | Fix |
|---------|-------|-----|
| kitten @ ls fails with "remote control not allowed" | allow_remote_control no in kitty.conf | Set to yes, reload config |
| LaunchAgent loads but last exit code = 78 | Script not executable | chmod +x ~/.local/bin/kitty-snapshot |
| LaunchAgent loads but never fires | StartInterval missing or zero | Set to a positive integer (seconds) |
| All snapshots empty [] | No kitty windows running, or wrong socket path | Check listen_on path matches --to |
| bootstrap fails with "already loaded" | Old load form still active | launchctl bootout first, then bootstrap |
| Restore opens windows with wrong shell | cmdline[0] is the shell, not the user command | Use cwd only and let the user re-run |
tools
Scaffold a new ComfyUI custom-node repo (pyproject, CI, release-please, vitest+pytest, JS extension skeleton) in the picker/gesture vein. Use when bootstrapping or init-ing a comfyui node pack.
tools
Orchestrate a ComfyUI node pack from idea to registry: scaffold, create + seed the repo, open the gitops adoption PR. Use when releasing or spinning up a new comfyui node pack.
testing
macOS EndpointSecurity/EDR high CPU & battery drain. Use when Kandji ESF / XProtect pegs a core; trace the exec storm via powermetrics + eslogger.
development
odiff pixel-by-pixel image diffing. Use when comparing screenshots, detecting visual regressions, diffing before/after PNGs, asserting golden images.