refactor(court-fetch): usage_status קורא ממקור-האמת המשותף usage_limits (איחוד ההעתק השלישי) (#268)

Co-authored-by: Chaim <chaim@marcus-law.co.il>
Co-committed-by: Chaim <chaim@marcus-law.co.il>

mcp-server/src/legal_mcp/court_fetch_service/server.py

@@ -43,6 +43,7 @@ if _pkg_root not in sys.path:
     sys.path.insert(0, _pkg_root)
 
 from legal_mcp.court_fetch_service import camofox_client  # noqa: E402
+from legal_mcp.services import usage_limits  # noqa: E402
 
 logger = logging.getLogger("legal_court_fetch_service")
 
@@ -95,14 +96,6 @@ async def _pm2_run(*args: str, timeout: float = 10) -> tuple[int, bytes, bytes]:
     return proc.returncode or 0, out, err
 
 
-# claude.ai subscription usage — the 5-hour / weekly utilization % the Claude
-# Code status bar shows, from the (undocumented) OAuth usage endpoint. Host-only:
-# the OAuth token lives in the CLI credentials file on the host, never in the
-# container. Read-only (no auth), like /pm2. The claude-code User-Agent is
-# REQUIRED — without it the request lands in an aggressively rate-limited bucket.
-_CLAUDE_CRED_PATH = "/home/chaim/.claude/.credentials.json"
-_OAUTH_USAGE_URL = "https://api.anthropic.com/api/oauth/usage"
-_USAGE_UA = "claude-code/2.1.177"
 # /operations polls every 5s; the usage endpoint 429s if hit that often (it's
 # meant for a status bar, not a poll loop). Cache the last good payload and only
 # re-fetch when older than this — Anthropic sees ~1 req/min regardless of how
@@ -115,35 +108,26 @@ async def usage_status(request: web.Request) -> web.Response:
     """Proxy the claude.ai subscription usage % (host-only — needs the local
     OAuth token), cached for _USAGE_TTL_SEC. On a fetch failure (e.g. the
     endpoint's own 429) serve the last good payload if we have one, so a
-    transient limit doesn't blank the dashboard."""
+    transient limit doesn't blank the dashboard.
+
+    The raw OAuth read is the SHARED single source of truth
+    (legal_mcp.services.usage_limits.subscription_usage) — the SAME reader the
+    halacha drain + supervisor gate on (G1/G2; no triplicated endpoint/creds/UA
+    constants). It's synchronous urllib, so run it in a thread to keep the aiohttp
+    event loop responsive."""
     now = time.monotonic()
     if _usage_cache["data"] is not None and (now - _usage_cache["ts"]) < _USAGE_TTL_SEC:
         return web.json_response(_usage_cache["data"])
 
-    try:
-        with open(_CLAUDE_CRED_PATH) as f:
-            token = json.load(f)["claudeAiOauth"]["accessToken"]
-    except Exception as e:
+    import asyncio as _asyncio
+    # subscription_usage returns None on ANY failure (creds missing / endpoint
+    # 429 / network) — it never throws; serve stale if we have it.
+    data = await _asyncio.get_event_loop().run_in_executor(
+        None, usage_limits.subscription_usage)
+    if data is None:
         if _usage_cache["data"] is not None:
             return web.json_response(_usage_cache["data"])
-        return web.json_response({"error": f"no claude credentials: {e}"}, status=502)
-
-    headers = {
-        "Authorization": f"Bearer {token}",
-        "User-Agent": _USAGE_UA,
-        "anthropic-beta": "oauth-2025-04-20",
-    }
-    try:
-        timeout = aiohttp.ClientTimeout(total=15)
-        async with aiohttp.ClientSession(timeout=timeout) as session:
-            async with session.get(_OAUTH_USAGE_URL, headers=headers) as r:
-                if r.status != 200:
-                    raise RuntimeError(f"usage endpoint {r.status}")
-                data = await r.json()
-    except Exception as e:  # never throw — serve stale if we have it
-        if _usage_cache["data"] is not None:
-            return web.json_response(_usage_cache["data"])
-        return web.json_response({"error": f"usage fetch failed: {e}"}, status=502)
+        return web.json_response({"error": "usage unavailable"}, status=502)
 
     _usage_cache["ts"] = now
     _usage_cache["data"] = data