docs/SPEC.md

@@ -435,8 +435,10 @@ Open questions on the schema:
     systemctl --user enable --now pkgwatch.timer
   ```
 
-  *`pkgwatch.service` sets `WorkingDirectory` to `~/dev/pkgwatch` because
-  `packages.d/`, `state/` and `work/` resolve relative to cwd.)*
+  *`pkgwatch.service` runs the release binary from the checkout and sets
+  no `WorkingDirectory`: config, state and work dirs come from the XDG
+  paths above, so the service needs `~/.config/pkgwatch/packages.d` set
+  up first — see the migration note under Paths.)*
 - **Notifications**: `notifier.rs` sends a desktop notification
   (`notify-send`) when a tier 4-6 release is newly queued for review or a
   tier 1-3 release is published; both are best-effort and never fail a

systemd/pkgwatch.service

@@ -1,9 +1,9 @@
 # User-level oneshot: one check -> fetch -> verify -> build -> publish pass.
 # Install: see docs/SPEC.md > Scheduling.
 #
-# WorkingDirectory matters: packages.d/, state/ and work/ are all resolved
-# relative to the cwd, so this must be the main checkout, not a worktree.
-# The binary is the release build in that same checkout (`cargo build
+# No WorkingDirectory: config, state and work dirs come from the XDG paths
+# in src/paths.rs (see docs/SPEC.md > Paths), not the cwd.
+# The binary is the release build in the main checkout (`cargo build
 # --release`), so a rebuild is what picks up code changes.
 [Unit]
 Description=pkgwatch: check tracked packages for new upstream releases
@@ -11,7 +11,6 @@ OnFailure=pkgwatch-failure.service
 
 [Service]
 Type=oneshot
-WorkingDirectory=%h/dev/pkgwatch
 ExecStart=%h/dev/pkgwatch/target/release/pkgwatch
 # Builds (makepkg, large Go/Rust binaries) can legitimately take a while.
 TimeoutStartSec=30min