Add qualifier prompt test page

src/vk_parser_app/admin.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import asyncio
 import json
 import base64
 import hashlib
@@ -26,6 +27,7 @@ from .db import fetch_setting, get_pool
 from .security import hash_password, new_token, token_hash, verify_password
 from .text_utils import normalize_hash_tag, parse_categories
 from .vk_api import VKAPIClient, normalize_vk_source
+from .workers.ai_qualifier import normalize_model, response_usage
 
 COOKIE_NAME = "vk_parser_admin"
 VK_OAUTH_VERIFIER_COOKIE = "vk_oauth_verifier"
@@ -914,6 +916,71 @@ async def category_rows() -> list[dict[str, Any]]:
     return result
 
 
+async def prompt_test_posts(selected_post_id: int | None = None) -> tuple[list[dict[str, Any]], dict[str, Any] | None]:
+    pool = await get_pool()
+    rows = await pool.fetch(
+        """
+        SELECT rp.id, rp.raw_text, rp.original_url, rp.posted_at, rp.created_at, rp.qualification_status,
+               rp.qualification_score, s.name AS source_name, s.tag AS source_tag,
+               (SELECT COUNT(*) FROM raw_post_media m WHERE m.raw_post_id=rp.id) AS media_count,
+               (SELECT ARRAY_AGG(DISTINCT m.media_type ORDER BY m.media_type)
+                FROM raw_post_media m WHERE m.raw_post_id=rp.id) AS media_types
+        FROM raw_posts rp
+        JOIN sources s ON s.id=rp.source_id
+        ORDER BY rp.created_at DESC, rp.id DESC
+        LIMIT 150
+        """
+    )
+    posts = []
+    for row in rows:
+        item = dict(row)
+        text = str(item.get("raw_text") or "").strip()
+        item["created_at_fmt"] = format_dt(item.get("created_at"))
+        item["posted_at_fmt"] = format_dt(item.get("posted_at"))
+        item["label"] = f"#{item['id']} · {item.get('source_name') or 'source'} · {item['created_at_fmt']}"
+        item["snippet"] = text[:180] + ("..." if len(text) > 180 else "")
+        posts.append(item)
+    selected = next((item for item in posts if int(item["id"]) == int(selected_post_id or 0)), None)
+    if not selected and selected_post_id:
+        row = await pool.fetchrow(
+            """
+            SELECT rp.id, rp.raw_text, rp.original_url, rp.posted_at, rp.created_at, rp.qualification_status,
+                   rp.qualification_score, s.name AS source_name, s.tag AS source_tag,
+                   (SELECT COUNT(*) FROM raw_post_media m WHERE m.raw_post_id=rp.id) AS media_count,
+                   (SELECT ARRAY_AGG(DISTINCT m.media_type ORDER BY m.media_type)
+                    FROM raw_post_media m WHERE m.raw_post_id=rp.id) AS media_types
+            FROM raw_posts rp
+            JOIN sources s ON s.id=rp.source_id
+            WHERE rp.id=$1
+            """,
+            selected_post_id,
+        )
+        if row:
+            selected = dict(row)
+            text = str(selected.get("raw_text") or "").strip()
+            selected["created_at_fmt"] = format_dt(selected.get("created_at"))
+            selected["posted_at_fmt"] = format_dt(selected.get("posted_at"))
+            selected["label"] = f"#{selected['id']} · {selected.get('source_name') or 'source'} · {selected['created_at_fmt']}"
+            selected["snippet"] = text[:180] + ("..." if len(text) > 180 else "")
+            posts.insert(0, selected)
+    if not selected and posts:
+        selected = posts[0]
+    return posts, selected
+
+
+def prompt_test_payload(post: dict[str, Any], max_text_chars: int) -> list[dict[str, Any]]:
+    return [
+        {
+            "id": int(post["id"]),
+            "source": post.get("source_name") or "",
+            "original_url": post.get("original_url") or "",
+            "media_count": int(post.get("media_count") or 0),
+            "media_types": list(post.get("media_types") or []),
+            "text": str(post.get("raw_text") or "")[:max_text_chars],
+        }
+    ]
+
+
 def setting_value(settings_rows: list[dict], key: str, default: Any = None) -> Any:
     for row in settings_rows:
         if row.get("key") == key:
@@ -2195,6 +2262,121 @@ async def raw_post_detail(request: Request, post_id: int):
     return templates.TemplateResponse("raw_post_detail.html", base_context(request, user, post=post))
 
 
+@app.get("/prompt-test", response_class=HTMLResponse)
+async def prompt_test(request: Request, post_id: int | None = None):
+    user = await get_current_user(request)
+    if not user:
+        return redirect("/login")
+    posts, selected = await prompt_test_posts(post_id)
+    max_text_chars = max(100, int(await fetch_setting("ai_qualifier_max_text_chars", 2000) or 2000))
+    prompt = str(await fetch_setting("ai_qualifier_prompt", "") or "")
+    provider = str(await fetch_setting("ai_qualifier_provider", "openrouter") or "openrouter")
+    model = str(await fetch_setting("ai_qualifier_model", "") or "")
+    payload = prompt_test_payload(selected, max_text_chars) if selected else []
+    return templates.TemplateResponse(
+        "prompt_test.html",
+        base_context(
+            request,
+            user,
+            posts=posts,
+            selected=selected,
+            prompt=prompt,
+            provider=provider,
+            model=model,
+            max_text_chars=max_text_chars,
+            payload_json=json.dumps(payload, ensure_ascii=False, indent=2),
+            result=None,
+            error=None,
+        ),
+    )
+
+
+@app.post("/prompt-test", response_class=HTMLResponse)
+async def prompt_test_run(
+    request: Request,
+    csrf_token: str = Form(...),
+    post_id: int = Form(...),
+    prompt: str = Form(...),
+):
+    user = await get_current_user(request)
+    if not user:
+        return redirect("/login")
+    require_csrf(user, csrf_token)
+    posts, selected = await prompt_test_posts(post_id)
+    max_text_chars = max(100, int(await fetch_setting("ai_qualifier_max_text_chars", 2000) or 2000))
+    provider = str(await fetch_setting("ai_qualifier_provider", "openrouter") or "openrouter")
+    model = str(await fetch_setting("ai_qualifier_model", "") or "").strip()
+    api_key = str(await fetch_setting("ai_qualifier_api_key", "") or "").strip()
+    api_base = str(await fetch_setting("ai_qualifier_api_base", "") or "").strip()
+    temperature = max(0.0, float(await fetch_setting("ai_qualifier_temperature", 0.0) or 0.0))
+    timeout = max(10, int(await fetch_setting("ai_qualifier_timeout_sec", 120) or 120))
+    normalized_prompt = prompt.replace("\\r\\n", "\n").replace("\\n", "\n").strip()
+    payload = prompt_test_payload(selected, max_text_chars) if selected else []
+    result = None
+    error = None
+    if not selected:
+        error = "Пост не найден."
+    elif not model or not api_key:
+        error = "Не заполнены модель или API-ключ квалификатора в настройках."
+    elif not normalized_prompt:
+        error = "Промпт пустой."
+    else:
+        try:
+            import litellm
+
+            kwargs: dict[str, Any] = {
+                "model": normalize_model(provider, model),
+                "messages": [
+                    {"role": "system", "content": normalized_prompt},
+                    {"role": "user", "content": json.dumps(payload, ensure_ascii=False)},
+                ],
+                "temperature": temperature,
+                "timeout": timeout,
+            }
+            if api_key:
+                kwargs["api_key"] = api_key
+            if api_base:
+                kwargs["api_base"] = api_base
+            response = await asyncio.to_thread(litellm.completion, **kwargs)
+            content = response.choices[0].message.content
+            parsed = None
+            raw_json = str(content or "").strip()
+            try:
+                if raw_json.startswith("```"):
+                    raw_json = raw_json.strip("`")
+                    if raw_json.lower().startswith("json"):
+                        raw_json = raw_json[4:].strip()
+                parsed = json.loads(raw_json)
+            except Exception:
+                parsed = None
+            result = {
+                "content": content,
+                "parsed_json": json.dumps(parsed, ensure_ascii=False, indent=2) if parsed is not None else "",
+                "usage": response_usage(response),
+                "model": kwargs["model"],
+            }
+            await audit(user["id"], "prompt_test.run", "raw_post", int(selected["id"]), {"provider": provider, "model": model})
+        except Exception as exc:
+            logger.exception("Prompt test failed")
+            error = str(exc)
+    return templates.TemplateResponse(
+        "prompt_test.html",
+        base_context(
+            request,
+            user,
+            posts=posts,
+            selected=selected,
+            prompt=normalized_prompt,
+            provider=provider,
+            model=model,
+            max_text_chars=max_text_chars,
+            payload_json=json.dumps(payload, ensure_ascii=False, indent=2),
+            result=result,
+            error=error,
+        ),
+    )
+
+
 @app.get("/workers", response_class=HTMLResponse)
 async def workers(request: Request):
     user = await get_current_user(request)

src/vk_parser_app/templates/base.html

@@ -117,6 +117,14 @@
     .prompt-hint { max-width:720px; border:1px solid var(--line); border-radius:8px; padding:10px 12px; background:#f8fafc; display:grid; gap:8px; color:var(--text); }
     .prompt-hint summary { cursor:pointer; font-weight:700; }
     .prompt-hint pre, .prompt-box { white-space:pre-wrap; overflow:auto; padding:10px; border:1px solid var(--line); border-radius:8px; background:#f8fafc; color:var(--text); }
+    .prompt-test-layout { display:grid; grid-template-columns:minmax(0, 1fr) 420px; gap:16px; align-items:start; }
+    .prompt-test-main { margin-bottom:0; }
+    .prompt-test-side { display:grid; gap:16px; position:sticky; top:74px; max-height:calc(100vh - 92px); overflow:auto; }
+    .prompt-test-model { min-height:38px; display:flex; align-items:center; gap:8px; flex-wrap:wrap; }
+    .prompt-test-post { margin:12px 0 14px; border:1px solid var(--line); border-radius:8px; padding:12px; background:var(--panel-soft); }
+    .prompt-test-post-head { display:flex; justify-content:space-between; gap:12px; align-items:center; margin-bottom:4px; }
+    .prompt-test-post pre { max-height:260px; overflow:auto; margin:10px 0 0; white-space:pre-wrap; font:inherit; color:var(--text); }
+    .prompt-test-help p { margin:0; }
     .hint-grid { display:grid; grid-template-columns:repeat(2, minmax(0,1fr)); gap:10px; }
     .hint-label { color:var(--muted); font-size:12px; font-weight:700; text-transform:uppercase; margin-bottom:3px; }
     .small-help { font-size:12px; margin-top:4px; }
@@ -208,7 +216,7 @@
     .regen-form { margin-top:16px; border-top:1px solid var(--line); padding-top:14px; }
     .compact-kv { margin-top:12px; grid-template-columns:90px minmax(0,1fr); }
     .empty-state { text-align:center; color:var(--muted); padding:34px; }
-    @media (max-width: 900px) { header { padding:8px 12px; align-items:flex-start; } .filter-grid, .sources-filter, .source-add-grid, .hint-grid, .advanced-grid, .tool-row, .tool-row.sort-row, .editor-filter, .page-head, .compare-grid, .dialog-grid, .catalog-layout, .media-upload-form, .category-create { grid-template-columns:1fr; } .page-head { display:grid; } .form-grid { grid-template-columns:1fr; } .detail-grid { grid-template-columns:1fr; } .editor-card { grid-template-columns:1fr; } .editor-actions { grid-template-columns:repeat(3, minmax(0,1fr)); } .context-pane { border-right:0; border-bottom:1px solid var(--line); max-height:none; } .edit-pane { max-height:none; } .side-panel, .filter-sidebar { position:static; max-height:none; } .list-toolbar { display:grid; align-items:start; } .compact-controls { justify-content:start; } main { padding:14px; } }
+    @media (max-width: 900px) { header { padding:8px 12px; align-items:flex-start; } .filter-grid, .sources-filter, .source-add-grid, .hint-grid, .advanced-grid, .tool-row, .tool-row.sort-row, .editor-filter, .page-head, .compare-grid, .dialog-grid, .catalog-layout, .media-upload-form, .category-create, .prompt-test-layout { grid-template-columns:1fr; } .page-head { display:grid; } .form-grid { grid-template-columns:1fr; } .detail-grid { grid-template-columns:1fr; } .editor-card { grid-template-columns:1fr; } .editor-actions { grid-template-columns:repeat(3, minmax(0,1fr)); } .context-pane { border-right:0; border-bottom:1px solid var(--line); max-height:none; } .edit-pane { max-height:none; } .side-panel, .filter-sidebar, .prompt-test-side { position:static; max-height:none; } .list-toolbar { display:grid; align-items:start; } .compact-controls { justify-content:start; } main { padding:14px; } }
   </style>
 </head>
 <body>
@@ -218,6 +226,7 @@
     <a href="/sources">Источники</a>
     <a href="/raw">Raw</a>
     <a href="/editor">Редакторская</a>
+    <a href="/prompt-test">Тест промптов</a>
     <a href="/workers">Воркеры</a>
     <a href="/users">Пользователи</a>
   </nav>

src/vk_parser_app/templates/prompt_test.html

@@ -0,0 +1,106 @@
+{% extends "base.html" %}
+{% block body %}
+<div class="page-head">
+  <div>
+    <h1>Тест промптов</h1>
+    <p class="muted">Песочница для свободной части <span class="mono">ai_qualifier_prompt</span>. Посты и оценки в БД не меняются.</p>
+  </div>
+  <a class="btn" href="/workers">Настройки AI</a>
+</div>
+
+<details class="panel prompt-test-help" open>
+  <summary><strong>Как пользоваться</strong></summary>
+  <div class="hint-grid">
+    <div>
+      <div class="hint-label">Что тестируем</div>
+      <p>Отправляется только текст из поля ниже как system prompt. <span class="mono">ai_qualifier_contract</span> специально не добавляется, чтобы можно было быстро проверять идеи, тон и критерии.</p>
+    </div>
+    <div>
+      <div class="hint-label">Что уйдёт в модель</div>
+      <p>Один выбранный raw-пост в том же формате, что использует квалификатор: <span class="mono">id</span>, источник, ссылка, количество медиа, типы медиа и обрезанный текст.</p>
+    </div>
+    <div>
+      <div class="hint-label">Что безопасно менять</div>
+      <p>Роль, критерии, примеры хороших/плохих постов, шкалу оценки и формулировки причин. Для рабочего воркера JSON-контракт всё равно живёт отдельно.</p>
+    </div>
+    <div>
+      <div class="hint-label">Что не происходит</div>
+      <p>Статус поста не меняется, batch не создаётся, оценка не сохраняется. Тест только тратит токены текущей модели квалификатора.</p>
+    </div>
+  </div>
+</details>
+
+<form class="prompt-test-layout" method="post" action="/prompt-test">
+  <input type="hidden" name="csrf_token" value="{{ user.csrf_token }}">
+  <section class="panel prompt-test-main">
+    <div class="form-grid">
+      <label>
+        <span>Пост для теста</span>
+        <select name="post_id" data-post-select>
+          {% for post in posts %}
+          <option value="{{ post.id }}" {% if selected and post.id == selected.id %}selected{% endif %}>{{ post.label }}</option>
+          {% endfor %}
+        </select>
+      </label>
+      <div>
+        <span class="muted">Модель</span>
+        <div class="prompt-test-model"><span class="pill">{{ provider }}</span><span class="mono">{{ model or "не выбрана" }}</span></div>
+      </div>
+    </div>
+
+    {% if selected %}
+    <div class="prompt-test-post">
+      <div class="prompt-test-post-head">
+        <strong>#{{ selected.id }} · {{ selected.source_name }}</strong>
+        <a href="{{ selected.original_url }}" target="_blank" rel="noopener">Оригинал</a>
+      </div>
+      <div class="muted">{{ selected.posted_at_fmt or selected.created_at_fmt }} · {{ selected.media_count }} медиа · {{ selected.media_types or [] }}</div>
+      <pre>{{ selected.raw_text }}</pre>
+    </div>
+    {% endif %}
+
+    <label>
+      <span>ai_qualifier_prompt для теста</span>
+      <textarea name="prompt" rows="18" spellcheck="false">{{ prompt }}</textarea>
+    </label>
+    <div class="dialog-actions">
+      <button class="btn primary" type="submit">Отправить тест</button>
+    </div>
+  </section>
+
+  <aside class="prompt-test-side">
+    <section class="panel">
+      <h2>Payload</h2>
+      <p class="muted">Именно это отправится user-сообщением. Текст обрезан по настройке <span class="mono">ai_qualifier_max_text_chars={{ max_text_chars }}</span>.</p>
+      <pre class="json-box">{{ payload_json }}</pre>
+    </section>
+
+    <section class="panel">
+      <h2>Ответ</h2>
+      {% if error %}
+      <div class="pill bad">{{ error }}</div>
+      {% elif result %}
+      <dl class="kv compact-kv">
+        <dt>Модель</dt><dd class="mono">{{ result.model }}</dd>
+        <dt>Токены</dt><dd>{{ result.usage.total_tokens or "—" }} total, {{ result.usage.prompt_tokens or "—" }} prompt, {{ result.usage.completion_tokens or "—" }} completion</dd>
+        <dt>Стоимость</dt><dd>{{ result.usage.estimated_cost_usd or "—" }}</dd>
+      </dl>
+      {% if result.parsed_json %}
+      <h3>JSON разобран</h3>
+      <pre class="json-box">{{ result.parsed_json }}</pre>
+      {% endif %}
+      <h3>Сырой ответ</h3>
+      <pre class="json-box">{{ result.content }}</pre>
+      {% else %}
+      <p class="muted">После отправки здесь появится ответ модели.</p>
+      {% endif %}
+    </section>
+  </aside>
+</form>
+<script>
+  document.querySelector("[data-post-select]")?.addEventListener("change", (event) => {
+    const id = event.target.value;
+    if (id) window.location.href = `/prompt-test?post_id=${encodeURIComponent(id)}`;
+  });
+</script>
+{% endblock %}