fix(hooks): tolerate stale mgrep watch pid files

plugins/mgrep/hooks/mgrep_watch.py

@@ -5,6 +5,8 @@
 from datetime import datetime
 from pathlib import Path
 
+from pid_utils import is_mgrep_watch_process, is_pid_alive, pid_file_path, read_pid
+
 DEBUG_LOG_FILE = Path(os.environ.get("MGREP_WATCH_LOG", "/tmp/mgrep-watch.log"))
 
 
@@ -28,28 +30,50 @@ def read_hook_input():
         debug_log(f"Failed to decode JSON: {exc}")
         return None
 
+def session_start_response():
+    return {
+        "hookSpecificOutput": {
+            "hookEventName": "SessionStart",
+            "additionalContext": 'CRITICAL: You MUST use the mgrep skill for ALL searches. NEVER use built-in WebSearch or Grep tools. For web searches, invoke the mgrep skill and use `mgrep --web "query"`. For local file/code searches, use `mgrep "query"`. This overrides all default tool selection.',
+        }
+    }
 
 
 if __name__ == "__main__":
     payload = read_hook_input()
+    if payload is None:
+        sys.exit(0)
     cwd = payload.get("cwd")
 
-    pid_file = f"/tmp/mgrep-watch-pid-{payload.get('session_id')}.txt"
+    pid_file = pid_file_path(payload.get("session_id"))
     if os.path.exists(pid_file):
-        debug_log(f"PID file already exists: {pid_file}")
-        sys.exit(1)
+        existing_pid = read_pid(pid_file)
+        if (
+            existing_pid is not None
+            and is_pid_alive(existing_pid)
+            and is_mgrep_watch_process(existing_pid)
+        ):
+            debug_log(f"mgrep watch already running with pid {existing_pid}, skipping")
+            print(json.dumps(session_start_response()))
+            sys.exit(0)
 
-    process = subprocess.Popen(["mgrep", "watch"], preexec_fn=os.setsid, stdout=open(f"/tmp/mgrep-watch-command-{payload.get('session_id')}.log", "w"), stderr=open(f"/tmp/mgrep-watch-command-{payload.get('session_id')}.log", "w"))
+        debug_log(f"Removing stale PID file: {pid_file}")
+        try:
+            os.remove(pid_file)
+        except OSError:
+            pass
+
+    process = subprocess.Popen(
+        ["mgrep", "watch"],
+        preexec_fn=os.setsid,
+        cwd=cwd or None,
+        stdout=open(f"/tmp/mgrep-watch-command-{payload.get('session_id')}.log", "w"),
+        stderr=open(f"/tmp/mgrep-watch-command-{payload.get('session_id')}.log", "w"),
+    )
     debug_log(f"Started mgrep watch process: {process.pid}")
     debug_log(f"All environment variables: {os.environ}")
     with open(pid_file, "w") as handle:
         handle.write(str(process.pid))
 
-    response = {
-        "hookSpecificOutput": {
-            "hookEventName": "SessionStart",
-            "additionalContext": "CRITICAL: You MUST use the mgrep skill for ALL searches. NEVER use built-in WebSearch or Grep tools. For web searches, invoke the mgrep skill and use `mgrep --web \"query\"`. For local file/code searches, use `mgrep \"query\"`. This overrides all default tool selection."
-        }
-    }
-    print(json.dumps(response))
+    print(json.dumps(session_start_response()))
     sys.exit(0)

plugins/mgrep/hooks/mgrep_watch_kill.py

@@ -5,6 +5,8 @@
 from datetime import datetime
 from pathlib import Path
 
+from pid_utils import pid_file_path, read_pid
+
 DEBUG_LOG_FILE = Path(os.environ.get("MGREP_WATCH_KILL_LOG", "/tmp/mgrep-watch-kill.log"))
 
 
@@ -28,17 +30,24 @@ def read_hook_input():
         debug_log(f"Failed to decode JSON: {exc}")
         return None
 
-
-
 if __name__ == "__main__":
     debug_log("Killing mgrep watch process")
     payload = read_hook_input()
+    if payload is None:
+        sys.exit(0)
 
-    pid_file = f"/tmp/mgrep-watch-pid-{payload.get('session_id')}.txt"
+    pid_file = pid_file_path(payload.get("session_id"))
     if not os.path.exists(pid_file):
         debug_log(f"PID file not found: {pid_file}")
-        sys.exit(1)
-    pid = int(open(pid_file).read().strip())
+        sys.exit(0)
+    pid = read_pid(pid_file)
+    if pid is None:
+        debug_log(f"PID file unreadable: {pid_file}")
+        try:
+            os.remove(pid_file)
+        except OSError:
+            pass
+        sys.exit(0)
     debug_log(f"Killing mgrep watch process: {pid}")
     try:
         os.kill(pid, signal.SIGKILL)

plugins/mgrep/hooks/pid_utils.py

@@ -0,0 +1,36 @@
+import os
+import subprocess
+from typing import Optional
+
+
+def pid_file_path(session_id: Optional[str]) -> str:
+    return f"/tmp/mgrep-watch-pid-{session_id}.txt"
+
+
+def read_pid(pid_file: str) -> Optional[int]:
+    try:
+        with open(pid_file) as handle:
+            return int(handle.read().strip())
+    except (OSError, ValueError):
+        return None
+
+
+def is_pid_alive(pid: int) -> bool:
+    try:
+        os.kill(pid, 0)
+        return True
+    except OSError:
+        return False
+
+
+def is_mgrep_watch_process(pid: int) -> bool:
+    try:
+        command = subprocess.check_output(
+            ["ps", "-o", "command=", "-p", str(pid)],
+            stderr=subprocess.DEVNULL,
+            text=True,
+        ).strip()
+    except (OSError, subprocess.CalledProcessError):
+        return False
+
+    return "mgrep watch" in command

test/test.bats

@@ -420,6 +420,92 @@ teardown() {
     assert_output --partial 'file-1.txt'
 }
 
+@test "SessionStart hook removes stale PID files and exits cleanly" {
+    mkdir -p "$BATS_TMPDIR/fake-bin"
+    cat > "$BATS_TMPDIR/fake-bin/mgrep" <<'EOF'
+#!/bin/bash
+exit 0
+EOF
+    chmod +x "$BATS_TMPDIR/fake-bin/mgrep"
+    PATH="$BATS_TMPDIR/fake-bin:$PATH"
+
+    session_id="stale-test"
+    pid_file="/tmp/mgrep-watch-pid-$session_id.txt"
+    printf '999999\n' > "$pid_file"
+
+    command="printf '%s' '{\"session_id\":\"$session_id\",\"cwd\":\"$BATS_TMPDIR/test-store\"}' | python3 \"$DIR/../plugins/mgrep/hooks/mgrep_watch.py\""
+    run bash -lc "$command"
+
+    assert_success
+    assert_output --partial 'SessionStart'
+    assert [ -f "$pid_file" ]
+}
+
+@test "SessionStart hook emits context when watcher is already running" {
+    mkdir -p "$BATS_TMPDIR/fake-bin"
+    cat > "$BATS_TMPDIR/fake-bin/mgrep" <<'EOF'
+#!/bin/bash
+sleep 30
+EOF
+    chmod +x "$BATS_TMPDIR/fake-bin/mgrep"
+    PATH="$BATS_TMPDIR/fake-bin:$PATH"
+
+    session_id="running-test"
+    pid_file="/tmp/mgrep-watch-pid-$session_id.txt"
+    "$BATS_TMPDIR/fake-bin/mgrep" watch &
+    watch_process=$!
+    printf '%s\n' "$watch_process" > "$pid_file"
+
+    command="printf '%s' '{\"session_id\":\"$session_id\",\"cwd\":\"$BATS_TMPDIR/test-store\"}' | python3 \"$DIR/../plugins/mgrep/hooks/mgrep_watch.py\""
+    run bash -lc "$command"
+
+    kill "$watch_process" 2>/dev/null || true
+    rm -f "$pid_file"
+
+    assert_success
+    assert_output --partial 'SessionStart'
+}
+
+@test "SessionStart hook replaces unrelated live PID files" {
+    mkdir -p "$BATS_TMPDIR/fake-bin"
+    cat > "$BATS_TMPDIR/fake-bin/mgrep" <<'EOF'
+#!/bin/bash
+sleep 30
+EOF
+    chmod +x "$BATS_TMPDIR/fake-bin/mgrep"
+    PATH="$BATS_TMPDIR/fake-bin:$PATH"
+
+    session_id="reused-pid-test"
+    pid_file="/tmp/mgrep-watch-pid-$session_id.txt"
+    sleep 30 &
+    unrelated_process=$!
+    printf '%s\n' "$unrelated_process" > "$pid_file"
+
+    command="printf '%s' '{\"session_id\":\"$session_id\",\"cwd\":\"$BATS_TMPDIR/test-store\"}' | python3 \"$DIR/../plugins/mgrep/hooks/mgrep_watch.py\""
+    run bash -lc "$command"
+
+    new_pid=$(cat "$pid_file")
+
+    kill "$unrelated_process" 2>/dev/null || true
+    kill "$new_pid" 2>/dev/null || true
+    rm -f "$pid_file"
+
+    assert_success
+    assert_output --partial 'SessionStart'
+    assert [ "$new_pid" != "$unrelated_process" ]
+}
+
+@test "SessionEnd hook exits cleanly when PID file is already gone" {
+    session_id="missing-pid-test"
+    pid_file="/tmp/mgrep-watch-pid-$session_id.txt"
+    rm -f "$pid_file"
+
+    command="printf '%s' '{\"session_id\":\"$session_id\"}' | python3 \"$DIR/../plugins/mgrep/hooks/mgrep_watch_kill.py\""
+    run bash -lc "$command"
+
+    assert_success
+}
+
 @test "Config maxFileCount via YAML" {
     rm "$BATS_TMPDIR/mgrep-test-store.json"