feat(agent-office): 큐레이터 파이프라인(fetch→claude→validate→save)

agent-office/app/curator/pipeline.py

@@ -0,0 +1,121 @@
+"""큐레이터 파이프라인 — fetch → claude → validate → save."""
+import json
+import time
+from typing import Any, Dict
+
+import httpx
+
+from ..config import ANTHROPIC_API_KEY, LOTTO_CURATOR_MODEL
+from .. import service_proxy
+from .prompt import SYSTEM_PROMPT, build_user_message
+from .schema import validate_response
+
+
+API_URL = "https://api.anthropic.com/v1/messages"
+
+
+class CuratorError(Exception):
+    pass
+
+
+async def _call_claude(user_text: str, feedback: str = "") -> tuple[dict, dict]:
+    if not ANTHROPIC_API_KEY:
+        raise CuratorError("ANTHROPIC_API_KEY missing")
+    headers = {
+        "x-api-key": ANTHROPIC_API_KEY,
+        "anthropic-version": "2023-06-01",
+        "anthropic-beta": "prompt-caching-2024-07-31",
+        "content-type": "application/json",
+    }
+    system_blocks = [{
+        "type": "text",
+        "text": SYSTEM_PROMPT,
+        "cache_control": {"type": "ephemeral"},
+    }]
+    if feedback:
+        user_text = f"이전 응답이 다음 이유로 거절됨: {feedback}\n올바른 스키마로 다시 응답.\n\n{user_text}"
+    payload = {
+        "model": LOTTO_CURATOR_MODEL,
+        "max_tokens": 4096,
+        "system": system_blocks,
+        "messages": [{"role": "user", "content": [{"type": "text", "text": user_text}]}],
+    }
+    started = time.monotonic()
+    async with httpx.AsyncClient(timeout=120) as client:
+        r = await client.post(API_URL, headers=headers, json=payload)
+        r.raise_for_status()
+        resp = r.json()
+    latency_ms = int((time.monotonic() - started) * 1000)
+
+    text = "".join(
+        b.get("text", "") for b in resp.get("content", []) if b.get("type") == "text"
+    ).strip()
+    if text.startswith("```"):
+        text = text.strip("`")
+        if text.startswith("json"):
+            text = text[4:]
+        text = text.strip()
+    parsed = json.loads(text)
+
+    usage = resp.get("usage", {}) or {}
+    return parsed, {
+        "input": int(usage.get("input_tokens", 0) or 0),
+        "output": int(usage.get("output_tokens", 0) or 0),
+        "cache_read": int(usage.get("cache_read_input_tokens", 0) or 0),
+        "cache_write": int(usage.get("cache_creation_input_tokens", 0) or 0),
+        "latency_ms": latency_ms,
+    }
+
+
+async def curate_weekly(source: str = "auto") -> Dict[str, Any]:
+    cand_resp = await service_proxy.lotto_candidates(n=20)
+    draw_no = cand_resp["draw_no"]
+    candidates = cand_resp["candidates"]
+    context = await service_proxy.lotto_context()
+
+    user_text = build_user_message(draw_no, candidates, {
+        "hot_numbers": context.get("hot_numbers", []),
+        "cold_numbers": context.get("cold_numbers", []),
+        "last_draw_summary": context.get("last_draw_summary", ""),
+        "my_recent_performance": context.get("my_recent_performance", []),
+    })
+
+    candidate_numbers = [c["numbers"] for c in candidates]
+
+    usage_total = {"input": 0, "output": 0, "cache_read": 0, "cache_write": 0, "latency_ms": 0}
+    last_error = None
+    validated = None
+
+    for attempt in (0, 1):
+        try:
+            raw, usage = await _call_claude(user_text, feedback=last_error or "")
+            for k in usage_total:
+                usage_total[k] += usage[k]
+            validated = validate_response(raw, candidate_numbers)
+            break
+        except Exception as e:
+            last_error = f"{type(e).__name__}: {e}"
+
+    if validated is None:
+        raise CuratorError(f"schema validation failed after retry: {last_error}")
+
+    payload = {
+        "draw_no": draw_no,
+        "picks": [p.model_dump() for p in validated.picks],
+        "narrative": validated.narrative.model_dump(),
+        "confidence": validated.confidence,
+        "model": LOTTO_CURATOR_MODEL,
+        "tokens_input": usage_total["input"],
+        "tokens_output": usage_total["output"],
+        "cache_read": usage_total["cache_read"],
+        "cache_write": usage_total["cache_write"],
+        "latency_ms": usage_total["latency_ms"],
+        "source": source,
+    }
+    await service_proxy.lotto_save_briefing(payload)
+    return {
+        "ok": True,
+        "draw_no": draw_no,
+        "confidence": validated.confidence,
+        "tokens": {"input": usage_total["input"], "output": usage_total["output"]},
+    }