fix: Dev-Agent handles all non-test issues, broaden issue scope beyond qe-feedback label

tests/acceptance/conftest.py

@@ -4,8 +4,11 @@ Usage::
 
     pytest tests/acceptance/ -v --run-acceptance [--acceptance-runs=3]
 
+LLM configuration is read from ``~/.openclaw/config/secrets.yaml``:
+    deepseek.apiKey / deepseek.baseUrl  → text model (deepseek-v4-flash)
+    dashscope.apiKey  / dashscope.baseUrl → vision model (qwen3-vl-plus)
+
 Environment variables:
-    DASHSCOPE_API_KEY  — LLM API key (required for Layers B/C)
     TEST_IR_PATH       — path to IR JSON to validate (default: ir_final.json sample)
     TEST_PARSED_PATH   — path to _parsed.json or _updated.json for coverage analysis
 """
@@ -20,17 +23,28 @@ from pathlib import Path
 from typing import Any
 
 import pytest
+import yaml
 
 # ── Path setup ──────────────────────────────────────────────────────────────
 
 _PROJECT_ROOT = Path(__file__).resolve().parent.parent.parent
 sys.path.insert(0, str(_PROJECT_ROOT))
 
+_SECRETS_PATH = Path.home() / ".openclaw" / "config" / "secrets.yaml"
+
 
 def _skill_path(skill_name: str) -> str:
     return str(_PROJECT_ROOT / "skills" / skill_name / "scripts")
 
 
+def _load_secrets() -> dict:
+    """Load LLM configuration from secrets.yaml."""
+    if _SECRETS_PATH.exists():
+        with open(_SECRETS_PATH, "r", encoding="utf-8") as f:
+            return yaml.safe_load(f) or {}
+    return {}
+
+
 # ── pytest configuration ────────────────────────────────────────────────────
 
 
@@ -77,11 +91,12 @@ def pytest_collection_modifyitems(config, items):
         skip_msg = pytest.mark.skip(reason="Need --run-acceptance flag to run")
         for item in acceptance_items:
             item.add_marker(skip_msg)
-        # Don't skip non-acceptance tests
         return
 
-    if not os.environ.get("DASHSCOPE_API_KEY"):
-        skip_msg = pytest.mark.skip(reason="DASHSCOPE_API_KEY not set")
+    secrets = _load_secrets()
+    has_api = bool(secrets.get("deepseek", {}).get("apiKey"))
+    if not has_api:
+        skip_msg = pytest.mark.skip(reason="No deepseek.apiKey in secrets.yaml")
         for item in acceptance_items:
             item.add_marker(skip_msg)
 
@@ -142,16 +157,93 @@ def parsed_data(parsed_path: str | None) -> dict | None:
         return json.load(f)
 
 
+# ── LLM client for acceptance tests ──────────────────────────────────────────
+
+
+class _AcceptanceLLM:
+    """Thin LLM wrapper for acceptance tests.
+
+    Uses deepseek-v4-flash for text (Layer C QE audit) via OpenAI-compatible API,
+    configured from ~/.openclaw/config/secrets.yaml.
+    """
+
+    TEXT_MODEL = "deepseek-v4-flash"
+    IMAGE_MODEL = "qwen3-vl-plus"
+    TIMEOUT = 180
+    MAX_RETRIES = 3
+
+    def __init__(self):
+        import time as _time
+        import openai
+
+        secrets = _load_secrets()
+        ds = secrets.get("deepseek", {})
+        ds_key = ds.get("apiKey", "") or os.environ.get("DEEPSEEK_API_KEY", "")
+        ds_base = ds.get("baseUrl", "https://api.deepseek.com/v1")
+
+        if not ds_key:
+            raise RuntimeError(
+                "No DeepSeek API key found. Set deepseek.apiKey in "
+                f"{_SECRETS_PATH} or DEEPSEEK_API_KEY env var."
+            )
+
+        self._api_key = ds_key
+        self._client = openai.OpenAI(
+            api_key=ds_key, base_url=ds_base, timeout=self.TIMEOUT, max_retries=self.MAX_RETRIES
+        )
+        self._prompt_tokens = 0
+        self._completion_tokens = 0
+        self._time = _time
+
+    def chat(self, model: str | None = None, messages: list[dict] | None = None,
+             response_format: dict | None = None) -> str:
+        """Send a chat completion request and return the text response."""
+        model = model or self.TEXT_MODEL
+        messages = messages or []
+
+        for attempt in range(self.MAX_RETRIES):
+            try:
+                kwargs = {"model": model, "messages": messages}
+                if response_format:
+                    kwargs["response_format"] = response_format
+                resp = self._client.chat.completions.create(**kwargs)
+                choice = resp.choices[0]
+                if choice.finish_reason == "length":
+                    raise RuntimeError(f"Response truncated (finish_reason=length)")
+                usage = resp.usage
+                if usage:
+                    self._prompt_tokens += usage.prompt_tokens or 0
+                    self._completion_tokens += usage.completion_tokens or 0
+                return choice.message.content or ""
+            except Exception as e:
+                if attempt < self.MAX_RETRIES - 1:
+                    delay = 2 ** attempt
+                    self._time.sleep(delay)
+                    continue
+                raise RuntimeError(f"LLM chat failed after {self.MAX_RETRIES} retries: {e}") from e
+        return ""
+
+    @property
+    def usage(self) -> dict:
+        return {
+            "prompt_tokens": self._prompt_tokens,
+            "completion_tokens": self._completion_tokens,
+            "total_tokens": self._prompt_tokens + self._completion_tokens,
+        }
+
+    @staticmethod
+    def estimate_tokens(text: str) -> int:
+        return max(1, len(text) // 3)
+
+
 @pytest.fixture(scope="session")
 def llm_client():
-    """Create an LLMClient instance for acceptance tests.
+    """Create an LLM client for acceptance tests.
 
-    Uses the DashScope-compatible LLMClient from the project.
+    Uses deepseek-v4-flash for text (Layer C QE audit), configured from
+    ~/.openclaw/config/secrets.yaml deepseek section.
     """
-    sys.path.insert(0, _skill_path("doc_parser_skill"))
-    from LLM import LLMClient
-
-    return LLMClient()
+    return _AcceptanceLLM()
 
 
 @pytest.fixture(scope="session")