fix(claw): warn if OpenClaw is running before migration

[fancydirty]

Detects running OpenClaw processes (via pgrep on Unix or tasklist on Windows) before hermes claw migrate and warns the user. Messaging platforms only allow one active session per bot token; running both simultaneously causes token fights and disconnects.

• Adds _is_openclaw_running() helper
• Adds _warn_if_openclaw_running() with a yes/no prompt (skippable with --yes)
• Covers Unix (pgrep) and Windows (tasklist) detection
• Includes tests for detection logic and warning behavior

Fixes #7907

[Copilot]

Pull request overview

Adds a pre-migration safety warning to hermes claw migrate by detecting whether OpenClaw appears to still be running, to reduce bot-token session conflicts during migration (Issue #7907).

Changes:

• Introduces _is_openclaw_running() to detect OpenClaw processes (pgrep on Unix, tasklist on Windows).
• Adds _warn_if_openclaw_running() that warns and prompts to continue (skippable with --yes) and integrates it into the migrate flow.
• Adds unit tests covering detection and warning behavior, plus a fixture to prevent prompts during existing migrate tests.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
hermes_cli/claw.py	Adds OpenClaw-running detection + warning prompt and calls it before gateway checks in migrate.
tests/hermes_cli/test_claw.py	Adds tests for detection/warning helpers and an autouse fixture to avoid interactive prompts in migrate tests.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Windows detection likely won’t work as written: tasklist output does not include process command lines (only the image name, PID, etc.), so searching stdout for openclaw/clawd will almost always return false even when OpenClaw is running under node.exe. Consider querying command lines via PowerShell/WMI (e.g., Get-CimInstance Win32_Process filtering on node.exe and inspecting CommandLine) or checking for an actual clawd.exe/openclaw*.exe image name if applicable, and update the tests accordingly.

Suggested change

	try:
	result = subprocess.run(
	["tasklist", "/FI", "IMAGENAME eq node.exe"],
	capture_output=True, text=True, timeout=5
	)
	output = result.stdout.lower()
	return "openclaw" in output or "clawd" in output
	except Exception:
	return False
	powershell_cmd = [
	"powershell",
	"-NoProfile",
	"-Command",
	(
	"$procs = Get-CimInstance Win32_Process | Where-Object { "
	"$_.Name -eq 'node.exe' -or "
	"$_.Name -eq 'clawd.exe' -or "
	"$_.Name -like 'openclaw*.exe' "
	"}; "
	"$procs | ForEach-Object { ($_.Name + ' ' + $_.CommandLine) }"
	),
	]
	try:
	result = subprocess.run(
	powershell_cmd,
	capture_output=True,
	text=True,
	timeout=5,
	)
	output = (result.stdout or "").lower()
	if "openclaw" in output or "clawd" in output:
	return True
	except (FileNotFoundError, subprocess.TimeoutExpired, OSError):
	pass
	for image_name in ("clawd.exe", "openclaw.exe"):
	try:
	result = subprocess.run(
	["tasklist", "/FI", f"IMAGENAME eq {image_name}"],
	capture_output=True,
	text=True,
	timeout=5,
	)
	if image_name in (result.stdout or "").lower():
	return True
	except (FileNotFoundError, subprocess.TimeoutExpired, OSError):
	continue
	return False

[Copilot]

_warn_if_openclaw_running() can prompt before the later non-interactive guard in _cmd_migrate (the sys.stdin.isatty() check). In a non-interactive session without --yes, prompt_yes_no() will hit EOF and exit(1), preventing the intended “preview only” behavior. Consider short-circuiting here when stdin isn’t a TTY (e.g., print the warning and return, or treat non-interactive as auto-yes/auto-no consistently with the rest of the command).

[Copilot]

These Windows tests assume tasklist returns command-line details like "node.exe openclaw-gateway", but tasklist output typically does not include the process command line. As a result, the test suite would validate behavior that can’t happen on real Windows, and would miss regressions once Windows detection is implemented correctly (e.g., via PowerShell/WMI). Update the tests to match the chosen Windows detection mechanism and real tasklist output formats.

[teknium1]

Merged via PR #8663. Your OpenClaw detection function and comprehensive tests were cherry-picked onto current main with your authorship preserved. Combined with PR #8555's systemd check and cleanup-side warning into a unified solution. Thanks @fancydirty!