Fix macOS DeepSeek Harness errors: launchd PATH and exFAT install failures

Two typical DeepSeek Harness errors on macOS: a launchd autostart crash loop (launchd does not inherit the shell PATH and cannot find node) and exFAT volume install failures (exFAT has no inode ownership checks, so pnpm errors during dependency installs). Fix them by writing a full PATH into the plist and moving installs onto APFS or a local volume, then verify.

Two symptoms: DeepSeek Harness autostart crash loop and exFAT install failures

The two symptoms differ: an autostart crash is a process-management problem, and an install failure is a filesystem problem. One by one (source: dshbase troubleshooting):

1. launchd cannot find node, autostart crash loop — after enabling autostart, the service crashes on launch and restarts repeatedly; running the same command manually in a terminal works fine;
2. exFAT volume install failure — installing with the project on an external exFAT volume (or a portable drive) makes pnpm fail (e.g. lefthook-related inode validation errors);
3. Manual runs fine but only autostart crashes → launchd; only a specific volume fails to install → filesystem.

DeepSeek Harness error roots: launchd does not inherit PATH, exFAT lacks inode checks

Two root causes: services started by launchd carry no shell PATH, and exFAT has no inode ownership checks so pnpm dependency installs fail validation. In detail:

1. launchd does not inherit the shell PATH — launchd is a system-level process manager; services it starts do not go through the shell and do not inherit the PATH from .zshrc; node installed in non-default directories such as /opt/homebrew/bin cannot be found (source: Apple launchd documentation);
2. exFAT has no inode ownership checks — exFAT is designed for removable storage and lacks POSIX inode ownership/permission metadata, so pnpm fails validation while installing dependencies (source: dshbase troubleshooting);
3. Both have corresponding discussions and fixes in the official repository (source: DeepSeek Harness issues).

Fix DeepSeek Harness errors: full PATH in the plist, install on APFS/local volume

Route by symptom: an autostart crash is fixed with plist environment variables, and an install failure by moving to APFS or a local volume. Step by step:

1. Find node's actual path (for the plist):
   bash

   which node
   # e.g. /opt/homebrew/bin/node (Homebrew on Apple Silicon)
   

2. Write a full PATH into the launchd plist — edit the plist and add EnvironmentVariables inside the dict:
   xml

   <key>EnvironmentVariables</key>
   <dict>
     <key>PATH</key>
     <string>/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin</string>
   </dict>
   

   Save and reload:
   bash

   launchctl unload <plist-path>
   launchctl load <plist-path>
   # Newer systems can use the equivalent bootout/bootstrap commands
   

   Confirm the PATH took effect with launchctl print <label> (source: Apple launchd documentation).
3. Install on APFS or a local volume (fixes exFAT install failures):
   • Copy the project from the external exFAT volume to the system disk (APFS) or a local volume;
   • Delete leftover node_modules and .pnpm-store on the volume to avoid stale metadata;
   • Re-run the install at the APFS path (pnpm install or dsh plugin ... add).
4. Verify:
   • Autostart: after a reboot the service starts on its own without the crash loop;
   • Install: pnpm install and dependency installs all succeed, and dsh web starts normally.

How to verify a DeepSeek Harness fix on macOS: PATH takes effect, autostart recovers, install completes

Three checks confirm the fix: the plist PATH took effect, autostart no longer crash-loops, and the install on APFS is all green — only when all three pass is it fixed. In order:

1. Confirm the plist PATH took effect — inspect the environment variables in launchd:

   bash

   launchctl print <label> | grep PATH
   

   Seeing /opt/homebrew/bin (which contains node) in the output means it took effect; only the default PATH means the plist is wrong — back to step 2 of the fix section.

2. Autostart regression check — start the service manually and confirm there is no crash loop:

   bash

   launchctl kickstart -k gui/$(id -u)/<label>
   

   The service keeps running and the process does not exit and restart repeatedly; if possible, reboot and confirm the service starts on its own.

3. Confirm the project sits on APFS or a local volume — check the filesystem type of the volume holding the working directory:

   bash

   mount | grep "<volume-path-or-working-dir>"
   # Expect apfs, hfs+ or another local format; exfat means keep migrating
   

4. Reinstall and verify — run the install under the APFS path:

   bash

   pnpm install
   

   All dependencies install cleanly with no inode/lefthook validation errors.

5. Full-chain verification — dsh web starts normally, a chat message gets a reply, and the autostart service runs after a reboot — the macOS platform issue is closed.

Notes: tell whether DeepSeek Harness has a launchd or filesystem issue

1. If manual runs work and only autostart crashes, check launchd's PATH first — do not doubt the installation itself.
2. Use the real path from which node when writing the plist PATH; different install methods give different paths.
3. External exFAT drives are fine for data backups, not for dependencies; install dependencies on APFS or a local volume.
4. Clear leftover node_modules before reinstalling after switching volumes, so stale filesystem metadata does not interfere.
5. See install error troubleshooting for other install issues.

Sources: dshbase troubleshooting, Apple developer documentation (launchd.plist), DeepSeek Harness official repository