fix: soft-fail model discovery timeouts with retries

OpenRouter /models list timed out and aborted the whole review.
Retry discovery, and if it still fails skip name validation so
parallel review can proceed.

scripts/resolve-providers.py

@@ -118,16 +118,19 @@ def secret_value(key_name: str) -> str:
     return ""
 
 
-def http_json(url: str, headers: dict) -> dict | list:
+def http_json(url: str, headers: dict, timeout: float = 25) -> dict | list:
     req = urllib.request.Request(url, headers=headers, method="GET")
-    with urllib.request.urlopen(req, timeout=60) as resp:
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
         return json.loads(resp.read().decode("utf-8"))
 
 
 def normalize_protocol(p: str) -> str:
     p = (p or "").strip().lower()
     if p not in PROTOCOL_WIRE:
-        die(f"unknown protocol {p!r} (want {' | '.join(PROTOCOLS)})")
+        die(
+            f"unknown protocol {p!r} (want {' | '.join(PROTOCOLS)})",
+            msg_zh=f"未知 protocol {p!r}（可选 {' | '.join(PROTOCOLS)}）",
+        )
     return p
 
 
@@ -135,41 +138,72 @@ def wire_protocol(p: str) -> str:
     return PROTOCOL_WIRE[normalize_protocol(p)]
 
 
-def discover_model_ids(base_url: str, protocol: str, api_key: str) -> set[str]:
+def discover_model_ids(base_url: str, protocol: str, api_key: str) -> set[str] | None:
+    """Return model id set, or None if discovery failed after retries (skip validation)."""
+    import time
+
     base = base_url.rstrip("/")
     wire = wire_protocol(protocol)
-    ids: set[str] = set()
+    if wire not in ("openai-completions", "openai-responses", "anthropic"):
-    try:
+        die(f"unhandled protocol wire {wire!r}")
-        if wire in ("openai-completions", "openai-responses"):
+
-            data = http_json(
+    last_err: Exception | None = None
-                f"{base}/models",
+    attempts = int(os.environ.get("GATEKEEP_DISCOVER_RETRIES", "3"))
-                {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
+    for attempt in range(max(1, attempts)):
+        try:
+            ids: set[str] = set()
+            if wire in ("openai-completions", "openai-responses"):
+                data = http_json(
+                    f"{base}/models",
+                    {
+                        "Authorization": f"Bearer {api_key}",
+                        "Content-Type": "application/json",
+                    },
+                )
+                for m in data.get("data") or []:
+                    mid = m.get("id")
+                    if mid:
+                        ids.add(mid)
+            else:
+                data = http_json(
+                    f"{base}/v1/models",
+                    {
+                        "x-api-key": api_key,
+                        "anthropic-version": "2023-06-01",
+                        "Content-Type": "application/json",
+                    },
+                )
+                for m in data.get("data") or []:
+                    mid = m.get("id")
+                    if mid:
+                        ids.add(mid)
+            return ids
+        except urllib.error.HTTPError as e:
+            body = e.read().decode("utf-8", errors="replace")[:400]
+            last_err = RuntimeError(f"HTTP {e.code} {body}")
+            # 401/403 are configuration errors — fail closed
+            if e.code in (401, 403):
+                die(
+                    f"model discovery failed for {base}: {last_err}",
+                    msg_zh=f"拉取模型列表失败（{base}）：{last_err} — 请检查 key",
+                )
+        except Exception as e:
+            last_err = e
+        if attempt + 1 < attempts:
+            sleep_s = 2**attempt
+            print(
+                f"gatekeep: model discovery retry {attempt + 1}/{attempts} "
+                f"after {sleep_s}s ({last_err})",
+                file=sys.stderr,
             )
-            for m in data.get("data") or []:
+            time.sleep(sleep_s)
-                mid = m.get("id")
+
-                if mid:
+    print(
-                    ids.add(mid)
+        f"gatekeep: model discovery failed for {base} after {attempts} tries "
-        elif wire == "anthropic":
+        f"({last_err}); skipping name validation and continuing",
-            data = http_json(
+        file=sys.stderr,
-                f"{base}/v1/models",
+    )
-                {
+    return None
-                    "x-api-key": api_key,
-                    "anthropic-version": "2023-06-01",
-                    "Content-Type": "application/json",
-                },
-            )
-            for m in data.get("data") or []:
-                mid = m.get("id")
-                if mid:
-                    ids.add(mid)
-        else:
-            die(f"unhandled protocol wire {wire!r}")
-    except urllib.error.HTTPError as e:
-        body = e.read().decode("utf-8", errors="replace")[:400]
-        die(f"model discovery failed for {base}: HTTP {e.code} {body}")
-    except Exception as e:
-        die(f"model discovery failed for {base}: {e}")
-    return ids
 
 
 def enrich_provider(pid: str, pconf: dict) -> dict:
@@ -198,6 +232,12 @@ def main() -> None:
     override_csv = sys.argv[3] if len(sys.argv) > 3 else ""
 
     cfg = load_raw(cfg_path)
+    # Resolve language early so die() can speak Chinese when configured
+    override_lang = (os.environ.get("GATEKEEP_LANGUAGE") or "").strip()
+    cfg_lang = str(cfg.get("language") or cfg.get("report_language") or "").strip()
+    language = override_lang or cfg_lang or "en"
+    os.environ["GATEKEEP_LANGUAGE"] = language
+
     providers_in = cfg["providers"]
     providers: dict[str, dict] = {
         pid: enrich_provider(pid, pconf) for pid, pconf in providers_in.items()
@@ -278,7 +318,7 @@ def main() -> None:
                 ),
             )
         available = discover_model_ids(pconf["base_url"], pconf["protocol"], api_key)
-        if not available:
+        if available is not None and len(available) == 0:
             die(
                 f"provider {pid}: model list empty — check base_url/protocol/key",
                 msg_zh=f"供应商 {pid} 模型列表为空 — 请检查 base_url / protocol / key",
@@ -289,14 +329,14 @@ def main() -> None:
             "wire_protocol": wire_protocol(pconf["protocol"]),
             "key_name": key_name,
             "api_key": api_key,
-            "available_count": len(available),
+            "available_count": len(available) if available is not None else -1,
-            "_available": available,
+            "_available": available,  # None ⇒ skip validation
         }
 
     def to_plan_entry(r: dict) -> dict:
         pid, model = r["provider"], r["model"]
         avail = provider_meta[pid]["_available"]
-        if model not in avail:
+        if avail is not None and model not in avail:
             sample = sorted(avail)[:8]
             die(
                 f"model {model!r} not available on provider {pid} "
@@ -324,11 +364,6 @@ def main() -> None:
     )
     os.chmod(secrets_path, 0o600)
 
-    # Action input GATEKEEP_LANGUAGE overrides YAML; empty → YAML; missing → en
-    override_lang = (os.environ.get("GATEKEEP_LANGUAGE") or "").strip()
-    cfg_lang = str(cfg.get("language") or cfg.get("report_language") or "").strip()
-    language = override_lang or cfg_lang or "en"
-
     plan = {
         "agreement_min": int(cfg.get("agreement_min", 2)),
         "temperature": cfg.get("temperature", 0),