refactor: AiServiceClient → thin AiGateway client (#126)

All AI calls now route through AiGateway (VPS):
- analyze task → LlmAdapter (sync, no repo needed)
- execute task → CodexAdapter (async, repo clone)

No API keys needed in this repo. Only CODEX_AUDIT_SERVICE_URL.

Co-authored-by: Claude <noreply@anthropic.com>

Author: Pigbibi

src/quant_platform_kit/strategy_lifecycle/ai_provider.py

@@ -1,31 +1,32 @@
-"""Unified AI Service Provider — two backend paths.
+"""Unified AI Service Provider — routes all calls through AiGateway.
 
-Architecture:
-  API calls (Claude/GPT) → direct API call (needs API_KEY secret)
-  Codex VPS calls        → CodexAuditBridge service (needs CODEX_AUDIT_SERVICE_URL)
+Architecture::
 
-The two paths exist because the VPS only runs Codex CLI. Claude/GPT are called
-directly from the GitHub Actions workflow using repo/org-level secrets.
+    QuantStrategyLifecycle          AiGateway (VPS, single service)
+    ───────────────────             ─────────────────────────────────
+    AiServiceClient                     │
+      ├─ review() ── task=analyze ──────┤──▶ LlmAdapter  (Claude/GPT API)
+      ├─ verify() ── task=execute ──────┤──▶ CodexAdapter (codex exec)
+      └─ execute()── task=execute ──────┤──▶ CodexAdapter (codex exec)
 
-Two built-in patterns:
+No API keys in this repo — all AI backends accessed through AiGateway.
+Only `CODEX_AUDIT_SERVICE_URL` is required.
 
-  RELIABILITY  — Codex VPS primary → Claude fallback → GPT fallback
-                 Used by: CodexAuditBridge (code audit, must complete).
-
-  SAFETY       — Rules → Claude → GPT → Codex VPS verify → consensus
-                 Used by: strategy_lifecycle (parameter approval, must be correct).
+Benefits:
+  - API keys live on the VPS (one place), not in N repos
+  - New backends = new adapter on the gateway, callers unchanged
+  - REPAIR/SAFETY patterns unchanged, just the transport is unified
 """
 
 from __future__ import annotations
 
 import enum
 import json
 import os
-import re
 import time
 import urllib.error
 import urllib.request
-from collections.abc import Mapping, Sequence
+from collections.abc import Sequence
 from dataclasses import dataclass
 from typing import Any
 
@@ -45,44 +46,30 @@ class AiPattern(str, enum.Enum):
 class AiProviderConfig:
     provider: AiProviderId
     label: str
-
-    # For API-based providers (Claude, GPT)
-    api_key_env: str = ""
-    model: str = ""
-    base_url: str = ""
-
-    # For Codex VPS
-    service_url_env: str = "CODEX_AUDIT_SERVICE_URL"
-    audience_env: str = "CODEX_AUDIT_SERVICE_AUDIENCE"
-    source_repo: str = "QuantStrategyLab/QuantStrategyLifecycle"
-    source_ref: str = "main"
-
+    model: str = ""                           # sent as "model" to gateway
+    task: str = "analyze"                     # analyze (API) or execute (Codex)
     can_execute_code: bool = False
     can_analyze: bool = True
 
-    def resolve_api_key(self) -> str | None:
-        if not self.api_key_env:
-            return None
-        return os.environ.get(self.api_key_env, "").strip() or None
-
     def resolve_service_url(self) -> str | None:
-        return os.environ.get(self.service_url_env, "").strip() or None
+        return os.environ.get("CODEX_AUDIT_SERVICE_URL", "").strip() or None
 
     @classmethod
     def claude(cls) -> "AiProviderConfig":
         return cls(provider=AiProviderId.CLAUDE, label="Claude",
-                   api_key_env="ANTHROPIC_API_KEY", model="claude-sonnet-4-6",
+                   model="claude-sonnet-4-6", task="analyze",
                    can_execute_code=False, can_analyze=True)
 
     @classmethod
     def gpt(cls) -> "AiProviderConfig":
         return cls(provider=AiProviderId.GPT, label="GPT",
-                   api_key_env="OPENAI_API_KEY", model="gpt-5.4-mini",
+                   model="gpt-5.4-mini", task="analyze",
                    can_execute_code=False, can_analyze=True)
 
     @classmethod
     def codex_vps(cls) -> "AiProviderConfig":
         return cls(provider=AiProviderId.CODEX_VPS, label="Codex VPS",
+                   task="execute",
                    can_execute_code=True, can_analyze=True)
 
 
@@ -93,52 +80,41 @@ class AiServiceConfig:
     fallback: tuple[AiProviderConfig, ...] = ()
     reviewers: tuple[AiProviderConfig, ...] = ()
     verifier: AiProviderConfig | None = None
-    require_consensus: bool = True
 
     @classmethod
     def reliability(cls, *, primary: AiProviderConfig, fallback: Sequence[AiProviderConfig] = ()) -> "AiServiceConfig":
         return cls(pattern=AiPattern.RELIABILITY, primary=primary, fallback=tuple(fallback))
 
     @classmethod
-    def safety(cls, *, reviewers: Sequence[AiProviderConfig], verifier: AiProviderConfig | None = None,
-               require_consensus: bool = True) -> "AiServiceConfig":
+    def safety(cls, *, reviewers: Sequence[AiProviderConfig], verifier: AiProviderConfig | None = None) -> "AiServiceConfig":
         return cls(pattern=AiPattern.SAFETY, reviewers=tuple(reviewers), verifier=verifier)
 
     @classmethod
     def from_env(cls) -> "AiServiceConfig":
-        """Auto-detect available backends from env vars."""
-        has_api_key = bool(os.environ.get("ANTHROPIC_API_KEY", "").strip())
-        has_gpt = bool(os.environ.get("OPENAI_API_KEY", "").strip())
-        has_codex = bool(os.environ.get("CODEX_AUDIT_SERVICE_URL", "").strip())
+        """Auto-detect from CODEX_AUDIT_SERVICE_URL (no API keys needed)."""
+        has_service = bool(os.environ.get("CODEX_AUDIT_SERVICE_URL", "").strip())
+        if not has_service:
+            return cls.safety(reviewers=[])
+        return cls.safety(
+            reviewers=[AiProviderConfig.claude(), AiProviderConfig.gpt()],
+            verifier=AiProviderConfig.codex_vps(),
+        )
 
-        reviewers: list[AiProviderConfig] = []
-        if has_api_key:
-            reviewers.append(AiProviderConfig.claude())
-        if has_gpt:
-            reviewers.append(AiProviderConfig.gpt())
 
-        verifier = AiProviderConfig.codex_vps() if has_codex else None
-
-        if reviewers:
-            return cls.safety(reviewers=reviewers, verifier=verifier)
-        if has_codex:
-            return cls.reliability(primary=AiProviderConfig.codex_vps())
-        return cls.safety(reviewers=[])
-
-
-# ── AI Service Client ────────────────────────────────────────────────
+# ── Client ───────────────────────────────────────────────────────────
 
 
 class AiServiceClient:
     def __init__(self, config: AiServiceConfig):
         self.config = config
 
-    def review(self, prompt: str, *, timeout: float = 45.0) -> list["AiCallResult"]:
+    def review(self, prompt: str, *, timeout: float = 120.0) -> list["AiCallResult"]:
+        """Run all reviewers (analyze/sync)."""
         import concurrent.futures
         if not self.config.reviewers:
             return []
         with concurrent.futures.ThreadPoolExecutor(max_workers=min(len(self.config.reviewers), 3)) as pool:
-            futures = {pool.submit(self._call_provider, c, prompt, timeout): c for c in self.config.reviewers}
+            futures = {pool.submit(self._call, c, prompt, timeout): c for c in self.config.reviewers}
             results = []
             for f in concurrent.futures.as_completed(futures):
                 try:
@@ -150,92 +126,70 @@ def review(self, prompt: str, *, timeout: float = 45.0) -> list["AiCallResult"]:
     def verify(self, prompt: str, *, timeout: float = 600.0) -> "AiCallResult | None":
         if self.config.verifier is None:
             return None
-        return self._call_provider(self.config.verifier, prompt, timeout)
+        return self._call(self.config.verifier, prompt, timeout)
 
     def execute(self, prompt: str, *, timeout: float = 600.0) -> "AiCallResult":
         if self.config.primary is not None:
-            r = self._call_provider(self.config.primary, prompt, timeout)
+            r = self._call(self.config.primary, prompt, timeout)
             if r.success:
                 return r
         for fb in self.config.fallback:
-            r = self._call_provider(fb, prompt, timeout)
+            r = self._call(fb, prompt, timeout)
             if r.success:
                 return AiCallResult(provider=r.provider, success=True, output=r.output, raw=r.raw,
-                                    note=f"Fallback after primary failed")
+                                    note="Fallback after primary failed")
         return AiCallResult.unavailable("all", "All providers exhausted")
 
-    # ── Core call routing ────────────────────────────────────────
-
-    def _call_provider(self, provider: AiProviderConfig, prompt: str, timeout: float) -> "AiCallResult":
-        if provider.provider == AiProviderId.CODEX_VPS:
-            return self._call_codex_vps(provider, prompt, timeout)
-        return self._call_llm_api(provider, prompt, timeout)
-
-    def _call_llm_api(self, provider: AiProviderConfig, prompt: str, timeout: float) -> "AiCallResult":
-        """Call Claude or GPT directly via API (uses repo secret)."""
-        api_key = provider.resolve_api_key()
-        if not api_key:
-            return AiCallResult.unavailable(provider.label, "API key not configured")
-
-        try:
-            from quant_strategy_plugins.ai_audit import AiAuditEndpoint, call_ai_audit
-
-            endpoint = AiAuditEndpoint(
-                name=f"ai_provider_{provider.provider.value}",
-                api_key=api_key,
-                provider="anthropic" if provider.provider == AiProviderId.CLAUDE else "openai",
-                model=provider.model,
-                base_url=provider.base_url or (
-                    "https://api.anthropic.com/v1" if provider.provider == AiProviderId.CLAUDE else "https://api.openai.com/v1"
-                ),
-            )
-            raw = call_ai_audit(endpoint, [{"role": "user", "content": prompt}], timeout=timeout)
-            output = raw if isinstance(raw, str) else json.dumps(raw)
-            return AiCallResult(provider=provider.label, success=True, output=output, raw=raw)
-        except ImportError:
-            return AiCallResult.unavailable(provider.label, "quant_strategy_plugins not installed")
-        except Exception as exc:
-            return AiCallResult.unavailable(provider.label, str(exc))
-
-    def _call_codex_vps(self, provider: AiProviderConfig, prompt: str, timeout: float) -> "AiCallResult":
-        """Call Codex VPS via the async job API."""
+    def _call(self, provider: AiProviderConfig, prompt: str, timeout: float) -> "AiCallResult":
+        """Call the AiGateway — all providers use the same endpoint."""
         service_url = provider.resolve_service_url()
         if not service_url:
             return AiCallResult.unavailable(provider.label, "CODEX_AUDIT_SERVICE_URL not configured")
 
         try:
             token = _fetch_oidc_token()
-            audience = os.environ.get(provider.audience_env, "quant-codex-audit")
             base_url = service_url.rstrip("/")
 
             payload = json.dumps({
-                "source_repository": provider.source_repo,
-                "source_ref": provider.source_ref,
-                "task": "strategy_review",
-                "mode": "review_only",
+                "task": provider.task,
+                "model": provider.model,
                 "prompt": prompt,
                 "timeout_seconds": int(timeout),
-                "model": "",  # service default model
+                "source_repository": os.environ.get("AI_GATEWAY_SOURCE_REPO", "QuantStrategyLab/QuantStrategyLifecycle"),
+                "source_ref": "main",
+                "mode": "review_only",
             }).encode("utf-8")
 
+            sync = provider.task == "analyze"
             req = urllib.request.Request(
                 f"{base_url}/v1/codex-audit/jobs", data=payload, method="POST",
                 headers={"Authorization": f"Bearer {token}", "Content-Type": "application/json",
-                         "Accept": "application/json", "User-Agent": "strategy-lifecycle"},
+                         "Accept": "application/json", "User-Agent": "quant-strategy-lifecycle"},
             )
             with urllib.request.urlopen(req, timeout=30) as resp:
-                submit = json.loads(resp.read().decode("utf-8"))
-            job_id = submit.get("job_id")
+                result = json.loads(resp.read().decode("utf-8"))
+
+            if sync:
+                # Analyze returns result inline
+                status = result.get("status")
+                if status == "succeeded":
+                    return AiCallResult(provider=provider.label, success=True,
+                                        output=str(result.get("output", "")), raw=result)
+                return AiCallResult(provider=provider.label, success=False,
+                                    output=result.get("error", "unknown"), raw=result)
+
+            # Execute returns async job_id → poll
+            job_id = result.get("job_id")
             if not isinstance(job_id, str) or not job_id:
-                return AiCallResult.unavailable(provider.label, "No job_id")
+                return AiCallResult.unavailable(provider.label, "No job_id from gateway")
 
             deadline = time.time() + timeout + 60
             while time.time() < deadline:
                 time.sleep(5)
                 req2 = urllib.request.Request(
                     f"{base_url}/v1/codex-audit/jobs/{job_id}", method="GET",
                     headers={"Authorization": f"Bearer {token}", "Accept": "application/json",
-                             "User-Agent": "strategy-lifecycle"},
+                             "User-Agent": "quant-strategy-lifecycle"},
                 )
                 try:
                     with urllib.request.urlopen(req2, timeout=30) as resp2:
@@ -246,7 +200,7 @@ def _call_codex_vps(self, provider: AiProviderConfig, prompt: str, timeout: floa
                 if status == "succeeded":
                     return AiCallResult(provider=provider.label, success=True,
                                         output=str(job.get("output", "")), raw=job)
-                elif status == "failed":
+                if status == "failed":
                     return AiCallResult(provider=provider.label, success=False,
                                         output=job.get("error", "unknown"), raw=job)
             return AiCallResult.unavailable(provider.label, "Timeout")
@@ -271,8 +225,10 @@ def _fetch_oidc_token(audience: str = "quant-codex-audit") -> str:
     token_url = os.environ.get("ACTIONS_ID_TOKEN_REQUEST_URL", "")
     token_bearer = os.environ.get("ACTIONS_ID_TOKEN_REQUEST_TOKEN", "")
     if token_url and token_bearer:
-        req = urllib.request.Request(f"{token_url}&audience={audience}",
-            headers={"Authorization": f"Bearer {token_bearer}"})
+        req = urllib.request.Request(
+            f"{token_url}&audience={audience}",
+            headers={"Authorization": f"Bearer {token_bearer}"},
+        )
         with urllib.request.urlopen(req, timeout=10) as resp:
             return str(json.loads(resp.read().decode("utf-8")).get("value", ""))
     return os.environ.get("CODEX_AUDIT_SERVICE_TOKEN", "").strip()