fix(core): don't mistake a stale daemon PID file for a running daemon (V2-1655)

.claude/skills/e2e-node-management-test/SKILL.md

@@ -104,6 +104,28 @@ ant node daemon start --json
 
 Re-run `ant node daemon info --json` and update the saved `api_base` URL (the port will have changed).
 
+**Step 5.5 — Stale PID file recovery:**
+
+A daemon that dies without cleaning up (a reboot, an OOM kill) leaves `daemon.pid` and
+`daemon.port` behind, and its PID can be reused by an unrelated process. Simulate that by
+recording the shell's own PID and a dead port as the daemon's. Run everything as **one** shell
+command so the shell (the "unrelated process") is still alive while `ant` runs:
+
+```bash
+# Linux:   DATA_DIR="${XDG_DATA_HOME:-$HOME/.local/share}/ant"
+# macOS:   DATA_DIR="$HOME/Library/Application Support/ant"
+# Windows: DATA_DIR="$APPDATA/ant", and write $(cat /proc/$$/winpid) instead of $$
+ant node daemon stop --json
+echo $$ > "$DATA_DIR/daemon.pid"; echo 1 > "$DATA_DIR/daemon.port"
+ant node daemon info --json; ant node daemon status --json; ant node daemon start --json
+```
+
+Verify: `info` and `status` both report `running: false`, and `start` reports
+`already_running: false` with a `port` that is present and not `1`. A stale port file must not be
+read back as the new daemon's port.
+
+Re-run `ant node daemon info --json` and update the saved `api_base` URL.
+
 ### Phase 6: Node management
 
 **Step 6.1 — Add nodes:**
@@ -516,6 +538,7 @@ Phase 5: Daemon Lifecycle
   [PASS] 5.2 Daemon status
   [PASS] 5.3 Daemon info
   [PASS] 5.4 Pinned-port flag round-trip
+  [PASS] 5.5 Stale PID file recovery
 
 Phase 6: Node Management
   [PASS] 6.1 Add 3 nodes

CHANGELOG.md

@@ -59,6 +59,25 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   metadata describing a file that was no longer there, so the next `npm update` silently rolled
   the user back.
 
+### Fixed
+- `ant node daemon info` and `ant node daemon start` no longer mistake a stale PID file for a
+  running daemon (V2-1655). A daemon that dies without cleaning up (a reboot, an OOM kill)
+  leaves its PID and port files behind, and once the PID is reused by an unrelated process,
+  `info` reported the dead daemon as running while `status` said it was not, and `start`
+  refused to start a new one. Both now require the process to be identifiable as the daemon,
+  the check `stop` already applied, and `start` removes the stale files before starting a new
+  daemon. On every platform the process must be `ant` running `ant node daemon run`, so another
+  `ant` command holding the reused PID is not taken for the daemon either. Where a process's
+  arguments or executable can't be read (another user's, an elevated or protected one on
+  Windows) or don't show that, it counts as the daemon only if the daemon's own API on the
+  recorded port reports its PID, on the configured listen address or either loopback (IPv4 or
+  IPv6), each asked with a one-second deadline. Otherwise `start` and `stop` say they can't tell
+  and leave the files alone, rather than risk orphaning a live daemon. Previously macOS trusted any
+  live PID and Windows checked only the executable's name, case-sensitively. `start` now fails
+  instead of carrying on when a stale file can't be removed, and it only spawns the daemon from
+  the `ant` binary. The `info` help text and README no longer claim its output is always JSON:
+  it is human-readable unless `--json` is passed, like every other command.
+
 ## [0.3.6] - 2026-09-01
 
 ### Fixed

README.md

@@ -523,7 +523,7 @@ Daemon is running
 
 #### `ant node daemon info`
 
-Output connection details as JSON (always JSON, regardless of `--json` flag). AI agents use this to discover the daemon's REST API.
+Show connection details: PID, port, console URL and REST API base. AI agents use `ant node daemon info --json` to discover the daemon's REST API.
 
 ```json
 {

ant-cli/src/commands/node/daemon.rs

@@ -48,7 +48,7 @@ pub enum DaemonCommand {
     Stop,
     /// Show whether the daemon is running and summary stats
     Status,
-    /// Output connection details for programmatic use (always JSON)
+    /// Show connection details: PID, port, console and API URLs (add --json for scripts)
     Info,
     /// Run the daemon in the foreground (used internally)
     #[command(hide = true)]