Make IgnoredKeywords.watch a real Link column to Settings

Replace the hand-copied search_url (URL type) with a proper relation:
- watch: LinkToAnotherRecord (belongs-to, many keywords -> one Settings row),
  picked from a dropdown in the UI so a wrong URL can't silently disable a keyword
- search_url: Lookup column that auto-fills Settings.url through the link

_ensure_keyword_watch_link() builds both columns idempotently on bootstrap and
upgrades a legacy URL-typed search_url to Link+Lookup. load_ignored_keywords()
now normalises Lookup values (string or array).

AGENTS.md

@@ -24,12 +24,13 @@ also archived but never alerted. Runs 24/7 as a Docker container on DSM (network
   monitor sets `skip_notify=true` + `notified=true` and does NOT send Telegram.
   The list is reloaded every notification cycle, so UI add/remove takes effect
   immediately.
-- Ignored keywords live in the NocoDB **IgnoredKeywords** table (`search_url` +
-  one `keyword` per row). A listing is silenced when its **title** contains any
-  keyword for its watch (matched by the watch's search URL), case-insensitive
-  substring match. Per-watch, not global: copy the `url` from the desired
-  `Settings` row into `search_url`. Reloaded every cycle, so UI edits take effect
-  immediately.
+- Ignored keywords live in the NocoDB **IgnoredKeywords** table: `watch` is a real
+  **Link column → `Settings`** (pick the watch from a dropdown) and `search_url` is a
+  **Lookup column** that auto-fills `Settings.url` from that link. A listing is
+  silenced when its **title** contains any keyword for its watch, case-insensitive
+  substring match. Per-watch, not global. Reloaded every cycle, so UI edits take
+  effect immediately. Bootstrap creates both columns and upgrades a legacy URL-typed
+  `search_url` to Link+Lookup automatically.
 - Dedupe key is `product_url` (`https://www.carousell.com.my/p/<id>/`), not the raw
   listing id and never the query-string URL.
 - First run per watch seeds the archive with **no** Telegram alert (`last_checked_at`

DOCUMENTATION.md

@@ -98,12 +98,18 @@ deleting a table is safe; it is recreated on the next start).
 
 | Column | Type | Purpose |
 |---|---|---|
-| `search_url` | URL | **copy the `url` from a `Settings` row** — which watch this keyword applies to |
+| `watch` | Link → `Settings` | **pick the watch from a dropdown** (belongs-to: many keywords → one watch) — no URL to copy by hand |
+| `search_url` | Lookup (`Settings.url`) | read-only, auto-filled from the linked watch — this is what the monitor reads |
 | `keyword` | SingleLineText | if the listing **title** contains this (case-insensitive substring), skip the Telegram alert (still archived) |
 
 One keyword per row; add multiple rows for multiple keywords. A keyword only
-silences listings found by the watch whose `Settings.url` matches `search_url` —
-the same keyword never applies to other watches. Empty rows are ignored.
+silences listings found by the watch you linked — the same keyword never applies to
+other watches. Rows with no watch or no keyword are ignored.
+
+`watch` is a real NocoDB Link column, so deleting/renaming a watch shows up in the
+relation, and there is no hand-copied URL that can silently drift out of sync.
+(The monitor's bootstrap also builds these two columns itself, including upgrading a
+legacy URL-typed `search_url` to Link+Lookup.)
 
 ### `Settings` (the watch list — you manage this)
 
@@ -150,7 +156,7 @@ the same keyword never applies to other watches. Empty rows are ignored.
 | Remove / pause a watch | Set `enabled` = false (or delete the row) |
 | Stop Telegram pings but keep archiving | Set `notify` = false |
 | Ignore a seller everywhere | Add `seller_name` row in `IgnoredSellers` |
-| Ignore certain keywords for **one watch** | Add row(s) in `IgnoredKeywords`: `search_url` = copy of that watch's `Settings.url`, `keyword` = e.g. `nike` (case-insensitive, matches inside the title) |
+| Ignore certain keywords for **one watch** | Add row(s) in `IgnoredKeywords`: pick the watch in the `watch` dropdown, `keyword` = e.g. `nike` (case-insensitive, matches inside the title). `search_url` fills itself |
 | Change how often it checks | Edit `check_interval_minutes` (5 = every 5 min) |
 | See what's new | Open `Listings`, sort by `first_seen_at` desc |
 | Browse with images | `Listings` grid view — the `image` column renders thumbnails |
@@ -230,8 +236,7 @@ Health file lives at `/data/health.json` inside the container:
 
 ## 11. Changelog
 
-- **2026-09-13** 新增 per-watch 忽略关键词：`IgnoredKeywords` 表（`search_url` +
-  `keyword`）。标题命中该 watch 关键词（大小写不敏感子串）时静默归档、不发
-  Telegram（`skip_notify=true`）。`search_url` 需与 `Settings.url` 一致，实现按
-  watch 隔离。
+- **2026-09-13** `IgnoredKeywords.watch` 改成真正的 NocoDB **Link 列**（Many-to-One → `Settings`），`search_url` 改成 **Lookup 列**自动带出 `Settings.url`（UI 下拉选 watch，不用手抄 URL）。bootstrap 会自建这两列，并把旧版 URL 型 `search_url` 自动升级成 Link+Lookup。
+- **2026-09-13** 新增 per-watch 忽略关键词：`IgnoredKeywords` 表。标题命中该 watch
+  关键词（大小写不敏感子串）时静默归档、不发 Telegram（`skip_notify=true`）。
 - **2026-09-08** 通知重构：每商品一条图文消息（title/price/condition/seller/url），归档与通知解耦（`notified` 列 + tick 末尾统一发 + 1s 间隔）。图片改用高清 URL（去 `_progressive_thumbnail`）。condition 归一化（New→Brand new、Used→Used，加第 6 档）。listed_at 加 `active_bump` fallback。修复 Telegram IPv6/DNS 问题（compose `extra_hosts` 钉 IPv4）。bot 换 `@carousellFoundBot`。

README.md

@@ -29,9 +29,11 @@ listing to a NocoDB base (with image URL + thumbnail), and alerts Telegram.
 **IgnoredSellers** — `seller_name`. Add/remove sellers here to suppress Telegram
 alerts for their listings (still archived, marked `skip_notify=true`).
 
-**IgnoredKeywords** — `search_url` + `keyword`. Per-watch title blocklist: keyword
-applies only to listings from the watch whose `Settings.url` matches `search_url`;
-case-insensitive substring match against the title. Still archived.
+**IgnoredKeywords** — `watch` (Link → Settings) + `keyword`. Per-watch title
+blocklist: pick the watch from a dropdown, add one keyword per row. A keyword only
+applies to listings from the linked watch; case-insensitive substring match against
+the title. `search_url` is a Lookup column that auto-fills from the linked watch.
+Still archived.
 
 ## Run
 

monitor.py

@@ -88,10 +88,10 @@ LISTINGS_COLS = [
 IGNORED_SELLERS_COLS = [
     ("seller_name", "SingleLineText"),
 ]
-# 忽略关键词（per-watch）：search_url=Settings 里的 watch URL 原样复制，keyword 单行一条。
+# 忽略关键词（per-watch）：keyword 单行一条；watch=Link 指向 Settings 记录；
+# search_url=Lookup 自动带出被链接 watch 的 Settings.url。
 # 大小写不敏感，命中该 watch 的「标题」即跳过通知（仍归档）。
 IGNORED_KEYWORDS_COLS = [
-    ("search_url", "URL"),
     ("keyword", "SingleLineText"),
 ]
 SETTINGS_COLS = [
@@ -247,9 +247,82 @@ def bootstrap():
     settings_tid = _ensure_table("Settings", SETTINGS_COLS)
     ignored_sellers_tid = _ensure_table("IgnoredSellers", IGNORED_SELLERS_COLS)
     ignored_keywords_tid = _ensure_table("IgnoredKeywords", IGNORED_KEYWORDS_COLS)
+    _ensure_keyword_watch_link(ignored_keywords_tid, settings_tid)
     return listings_tid, settings_tid, ignored_sellers_tid, ignored_keywords_tid
 
 
+def _ensure_keyword_watch_link(kw_tid, settings_tid):
+    """IgnoredKeywords 的 watch(Link→Settings) + search_url(Lookup→Settings.url)。
+
+    NocoDB 2026.08 的列创建 API 与旧版不同：
+      - Link 列: uidt="LinkToAnotherRecord", 顶层 parentId/childId/type
+      - 回读 link 列的 colOptions.fk_column_id（物理外键列）作为 Lookup 的
+        fk_relation_column_id
+      - Lookup 列: uidt="Lookup", 顶层 fk_relation_column_id + fk_lookup_column_id
+
+    幂等：列已存在就跳过。旧版遗留的 URL 型 search_url 会被替换成 Lookup。
+    """
+
+    def _kw_meta():
+        st, m = nc("GET", f"/api/v2/meta/tables/{kw_tid}")
+        if st != 200:
+            raise RuntimeError(f"read IgnoredKeywords meta failed: {m}")
+        return m
+
+    m = _kw_meta()
+    cols = {c.get("title"): c for c in m.get("columns", [])}
+
+    # 1. watch: LinkToAnotherRecord -> Settings
+    link_col = None
+    if "watch" in cols:
+        if cols["watch"].get("uidt") != "LinkToAnotherRecord":
+            nc("DELETE", f"/api/v2/meta/columns/{cols['watch']['id']}")
+        else:
+            link_col = cols["watch"]
+
+    if link_col is None:
+        st, out = nc("POST", f"/api/v2/meta/tables/{kw_tid}/columns",
+                     {"title": "watch", "uidt": "LinkToAnotherRecord",
+                      "parentId": settings_tid, "childId": kw_tid,
+                      "type": "bt", "colOptions": {}})
+        if st != 200:
+            raise RuntimeError(f"create watch link failed: {out}")
+        m = _kw_meta()
+        cols = {c.get("title"): c for c in m.get("columns", [])}
+        link_col = cols.get("watch")
+
+    if link_col is None:
+        raise RuntimeError("watch link column missing after create")
+
+    # 2. search_url: Lookup (auto-pull Settings.url through the link)
+    if "search_url" in cols and cols["search_url"].get("uidt") == "Lookup":
+        return  # already good
+
+    # drop stale non-Lookup search_url (legacy URL column)
+    if "search_url" in cols:
+        nc("DELETE", f"/api/v2/meta/columns/{cols['search_url']['id']}")
+
+    fk_parent = (link_col.get("colOptions") or {}).get("fk_column_id")
+    if not fk_parent:
+        raise RuntimeError("watch link has no fk_column_id")
+
+    st, sm = nc("GET", f"/api/v2/meta/tables/{settings_tid}")
+    if st != 200:
+        raise RuntimeError(f"read Settings meta failed: {sm}")
+    url_col = next((c for c in sm.get("columns", [])
+                    if c.get("title") == "url"), None)
+    if url_col is None:
+        raise RuntimeError("Settings.url column not found")
+
+    st, out = nc("POST", f"/api/v2/meta/tables/{kw_tid}/columns",
+                 {"title": "search_url", "uidt": "Lookup",
+                  "fk_relation_column_id": fk_parent,
+                  "fk_lookup_column_id": url_col["id"],
+                  "colOptions": {}})
+    if st != 200:
+        raise RuntimeError(f"create search_url lookup failed: {out}")
+
+
 # --------------------------------------------------------------------------- #
 # Carousell extraction
 # --------------------------------------------------------------------------- #
@@ -384,13 +457,18 @@ def load_ignored_keywords(ignored_keywords_tid):
     """从 IgnoredKeywords 表读取忽略关键词，返回 {search_url: {小写关键词}}。
 
     关键词按 watch（search_url）分组；匹配时大小写不敏感。
+    search_url 是 Lookup 列（经 watch 链接自动带出 Settings.url），
+    API 可能返回字符串或数组，这里统一归一化。
     """
     ignored = {}
     st, j = nc("GET", f"/api/v2/tables/{ignored_keywords_tid}/records?limit=1000")
     if st != 200:
         raise RuntimeError(f"load ignored keywords failed: {j}")
     for r in j.get("list", []):
-        url = (r.get("search_url") or "").strip()
+        url = r.get("search_url") or ""
+        if isinstance(url, list):  # Lookup 有时返回数组
+            url = url[0] if url else ""
+        url = str(url).strip()
         kw = (r.get("keyword") or "").strip().lower()
         if not url or not kw:
             continue