feat(sources): feishu pack real-workspace verification — tasks/wiki reconciled, has_more_path termination

Phase-4 live pass against a real Feishu workspace (OAuth user token,
gateway v1.3.3), which caught two contract-level defects the mock suite
could not see:

- Real task rows carry NO `completed` boolean — completion is
  `status` (todo|done) plus `completed_at`, whose "not completed"
  spelling is the digit string "0" (epoch-zero sentinel, documented at
  the column). The draft `completed` column and its Inexact pushdown
  are gone: the mapping would have re-trimmed every row to zero — the
  always-NULL class the live pass exists to catch. `status` is mapped
  instead; the action's `completed` input stays deliberately unmapped.
- Feishu's wiki space listing answers its FINAL page with
  `has_more: false` beside a NON-empty `page_token` ("0||…"), so
  null-token cursor termination refetches a finished scan and dies as
  PaginationLoop. New engine field `has_more_path` on the Cursor
  strategy (optional, backward-compatible): when declared it is the
  authoritative termination signal; `hasMore: true` without a usable
  token is PaginationCursorInvalid, a non-boolean signal is the new
  PaginationHasMoreInvalid — never a silent truncation. Declared on all
  six feishu tables; 4 engine tests + a live-shape e2e pin both sides.

Live evidence (skardi-server against the real gateway): registration
passes the fingerprint gate on LIVE discovery; tasks 9 rows (6 todo /
3 done) with EVERY mapped column non-NULL somewhere (due on 2 rows);
wiki_spaces completes on the real hasMore:false + non-empty-token page;
wiki_nodes decodes epoch-second strings to real instants. tasks /
wiki_spaces / wiki_nodes fixtures are now redacted live captures;
wiki_nodes gains creator/url, wiki_spaces gains open_sharing.

chats / messages / chat_members remain drafts: Feishu gates im APIs on
the app's bot capability (232025) independently of OAuth scopes; their
live pass lands when the test app has it enabled.

Full suite: cargo test -p skardi --lib → 827 passed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: bakey

crates/skardi/src/sources/providers/open_connector/error.rs

@@ -257,6 +257,19 @@ pub enum OpenConnectorError {
         found: String,
     },
 
+    /// The declared has-more signal was absent or not a boolean. Guessing
+    /// either way could truncate or loop the scan; carries the JSON *kind*
+    /// only, never the value.
+    #[error(
+        "Open Connector pagination has-more signal at '{path}' on page {page} is {found}, \
+         expected a boolean"
+    )]
+    PaginationHasMoreInvalid {
+        path: String,
+        page: usize,
+        found: String,
+    },
+
     /// Pagination failed to advance: the gateway returned an already-seen
     /// cursor, which would loop the scan forever.
     #[error(

crates/skardi/src/sources/providers/open_connector/packs/feishu.rs

@@ -21,12 +21,16 @@
 //!
 //! - **Cursor pagination on every table** (`pageToken` in, top-level
 //!   `$.pageToken` out; page size 100 — im/tasks maximum — or 50, the
-//!   wiki maximum). Termination is complete on the null-token spelling
-//!   the executors normalize to; `hasMore` is redundant with the null
-//!   token and deliberately unused. A repeated token fails as
-//!   `PaginationLoop`, a non-string token as `PaginationCursorInvalid` —
-//!   never a silent truncation. No feishu executor filters fetched pages,
-//!   so the cursor's termination signal is undamaged.
+//!   wiki maximum), with `$.hasMore` declared as the AUTHORITATIVE
+//!   termination signal (`has_more_path`). That is a live-verification
+//!   correction, not a nicety: Feishu's wiki space listing answers its
+//!   final page with `has_more: false` beside a NON-empty `page_token`
+//!   (`"0||…"`, captured 2026-08-04), so the null-token spelling alone
+//!   would refetch a finished scan and die as `PaginationLoop`. With the
+//!   signal declared, `hasMore: true` without a usable token is contract
+//!   drift (`PaginationCursorInvalid`), a non-boolean signal is
+//!   `PaginationHasMoreInvalid` — never a silent truncation. No feishu
+//!   executor filters fetched pages, so the signals are undamaged.
 //! - **Orderings are pinned for cursor stability**: `chats` and
 //!   `messages` pin `sortType: ByCreateTimeAsc` because the API's
 //!   activity-ordered default reshuffles rows mid-scan (skips and
@@ -41,9 +45,15 @@
 //!   `endTime` is deliberately unmapped: it is EXCLUSIVE, and flooring
 //!   an upper bound drops rows.
 //! - **`tasks` pins `type: my_tasks`** (the executor default, declared
-//!   rather than inherited) and omits `completed` by default — Feishu's
-//!   state=all. The `completed` equality pushdown stays `Inexact` like
-//!   every enum-ish push.
+//!   rather than inherited) and omits the action's `completed` input —
+//!   Feishu's state=all. Live rows (2026-08-04) carry NO `completed`
+//!   boolean: completion is `status` (todo|done) plus `completed_at`,
+//!   whose "not completed" spelling is the digit string "0" (epoch-zero
+//!   sentinel, documented at the column). The `completed` input is
+//!   therefore deliberately unmapped — a `status` string cannot render
+//!   into a boolean input without a value transform the filter engine
+//!   deliberately lacks, and a mapping on an always-NULL draft column
+//!   would have re-trimmed every row to zero.
 //! - **`wiki_nodes` lists ONE level** (children of `parentNodeToken`,
 //!   space root when omitted) — the action's own shape; full-tree
 //!   traversal is client-side recursion, documented in the pack doc.
@@ -55,11 +65,13 @@
 //!   item schema LOOSE (`additionalProperties: true`, zero declared
 //!   properties) — so ALL mapped columns ride passthrough outside the
 //!   fingerprint gate, and the coverage-gap pin records that honestly.
-//!   Column truth is therefore settled ONLY by real rows: the sets
-//!   below are drafted from Feishu's API reference and MUST be
-//!   reconciled against a live workspace (skill phase 4) before this
-//!   pack is submission-ready; the fixtures are synthetic drafts to be
-//!   re-derived as redacted live captures in that phase.
+//!   Column truth is therefore settled ONLY by real rows. `tasks`,
+//!   `wiki_spaces`, and `wiki_nodes` are reconciled against a live
+//!   workspace (2026-08-04; their fixtures are redacted live captures).
+//!   `chats` / `messages` / `chat_members` remain API-reference drafts
+//!   with synthetic fixtures: Feishu gates the im APIs on the app's BOT
+//!   capability (error 232025) independently of OAuth scopes, and the
+//!   live pass completes once the test app has it enabled.
 
 use std::sync::OnceLock;
 
@@ -210,35 +222,46 @@ mod tests {
     }
 
     #[test]
-    fn tasks_fixture_converts_booleans_and_empty_lists() {
+    fn tasks_fixture_converts_status_and_the_epoch_zero_sentinel() {
+        // Redacted live capture (2026-08-04): completion is `status`
+        // (todo|done) plus `completed_at` — there is NO `completed`
+        // boolean on the wire, which is why that draft column is gone.
         let batch = convert_fixture(table("tasks"), include_str!("fixtures/feishu/tasks.json"));
-        assert_eq!(batch.num_rows(), 2);
-        assert!(!boolean(&batch, "completed").value(0));
-        assert!(boolean(&batch, "completed").value(1));
-        assert!(utf8(&batch, "summary").is_null(1), "JSON null → SQL NULL");
-        // JSON null due is SQL NULL; the empty members array is the JSON
-        // text `[]`, not NULL — an empty list is data.
-        assert!(utf8(&batch, "due").is_null(1));
-        assert_eq!(utf8(&batch, "members").value(1), "[]");
+        assert_eq!(batch.num_rows(), 3);
+        assert_eq!(utf8(&batch, "status").value(0), "todo");
+        assert_eq!(utf8(&batch, "status").value(2), "done");
+        // Feishu spells "not completed" as "0" — the epoch-zero sentinel
+        // the YAML documents; a done task carries a real instant.
+        assert_eq!(millis(&batch, "completed_at").value(0), 0);
+        assert_eq!(millis(&batch, "completed_at").value(2), 1_780_322_529_638);
+        assert_eq!(millis(&batch, "created_at").value(0), 1_775_202_331_888);
+        // A present JSON null `due` is SQL NULL; members survive as
+        // opaque JSON.
+        assert!(utf8(&batch, "due").is_null(0));
+        let members: Value =
+            serde_json::from_str(utf8(&batch, "members").value(0)).expect("valid JSON");
+        assert!(members.is_array());
     }
 
     #[test]
     fn wiki_nodes_fixture_converts_epoch_second_strings() {
+        // Redacted live capture (2026-08-04).
         let batch = convert_fixture(
             table("wiki_nodes"),
             include_str!("fixtures/feishu/wiki_nodes.json"),
         );
-        assert_eq!(batch.num_rows(), 2);
+        assert_eq!(batch.num_rows(), 3);
         // Epoch-SECONDS digit strings scale to millis.
         assert_eq!(
             millis(&batch, "obj_create_time").value(0),
-            1_735_689_600_000
+            1_636_114_726_000
         );
-        assert!(millis(&batch, "obj_create_time").is_null(1));
-        assert!(boolean(&batch, "has_child").value(0));
+        assert!(!boolean(&batch, "has_child").value(0));
         // Feishu spells "no parent" as the empty string — that is data,
         // preserved verbatim, not coerced to NULL.
         assert_eq!(utf8(&batch, "parent_node_token").value(0), "");
+        assert!(utf8(&batch, "creator").value(0).starts_with("ou_"));
+        assert!(utf8(&batch, "url").value(0).starts_with("https://"));
     }
 
     #[test]
@@ -255,9 +278,13 @@ mod tests {
             table("wiki_spaces"),
             include_str!("fixtures/feishu/wiki_spaces.json"),
         );
-        assert_eq!(spaces.num_rows(), 2);
-        assert_eq!(utf8(&spaces, "space_id").value(0), "70001");
-        assert!(utf8(&spaces, "description").is_null(1));
+        // Redacted live capture (2026-08-04): note the envelope carries a
+        // NON-empty pageToken beside hasMore:false — the shape the
+        // has_more_path termination exists for, pinned end to end by
+        // `wiki_spaces_terminate_on_has_more_false_despite_a_token`.
+        assert_eq!(spaces.num_rows(), 1);
+        assert_eq!(utf8(&spaces, "space_id").value(0), "700000038");
+        assert_eq!(utf8(&spaces, "open_sharing").value(0), "closed");
     }
 
     #[test]
@@ -583,7 +610,11 @@ bindings:
     }
 
     #[tokio::test]
-    async fn tasks_pin_their_population_and_push_completed_equality() {
+    async fn tasks_pin_their_population_and_keep_status_predicates_local() {
+        // The action HAS a `completed` boolean input, but real rows carry
+        // `status` (todo|done) and no boolean — so nothing is mapped and a
+        // status predicate must run entirely in DataFusion while the
+        // request carries exactly the pins and pagination, nothing else.
         let gateway = MockGateway::start(|req| {
             if req.method == "GET" && req.path == "/v1/health" {
                 return MockResponse::ok("{}");
@@ -594,8 +625,8 @@ bindings:
             if req.method == "POST" && req.path == "/v1/actions/feishu.list_tasks" {
                 return MockResponse::ok(&envelope_ok(
                     &json!({"items": [
-                        {"guid": "t-1", "completed": true},
-                        {"guid": "t-2", "completed": false}],
+                        {"guid": "t-1", "status": "done"},
+                        {"guid": "t-2", "status": "todo"}],
                         "pageToken": null, "hasMore": false})
                     .to_string(),
                 ));
@@ -606,11 +637,7 @@ bindings:
         let (gateway, ctx) =
             setup_with_gateway(gateway, "SKARDI_TEST_OC_FEISHU_TASKS", "tasks").await;
 
-        let batches = collect(
-            &ctx,
-            "SELECT guid FROM saas.ws.tasks WHERE completed = true",
-        )
-        .await;
+        let batches = collect(&ctx, "SELECT guid FROM saas.ws.tasks WHERE status = 'done'").await;
         let guids: Vec<String> = batches
             .iter()
             .flat_map(|b| {
@@ -624,11 +651,59 @@ bindings:
                 (0..col.len()).map(move |i| col.value(i).to_string())
             })
             .collect();
-        assert_eq!(guids, vec!["t-1"], "Inexact push re-trimmed locally");
+        assert_eq!(guids, vec!["t-1"], "status filtering happened locally");
 
         let input = &execute_inputs(&gateway)[0];
         assert_eq!(input["type"], "my_tasks", "population pin");
-        assert_eq!(input["completed"], true, "boolean equality pushed");
+        let mut keys: Vec<&str> = input
+            .as_object()
+            .expect("input object")
+            .keys()
+            .map(String::as_str)
+            .collect();
+        keys.sort_unstable();
+        assert_eq!(
+            keys,
+            vec!["pageSize", "type"],
+            "the status predicate is not pushed and no `completed` input is invented"
+        );
+    }
+
+    #[tokio::test]
+    async fn wiki_spaces_terminate_on_has_more_false_despite_a_token() {
+        // The live wire shape this pack's has_more_path exists for
+        // (captured 2026-08-04): Feishu wiki answers the FINAL page with
+        // hasMore:false but a NON-empty pageToken ("0||…"). Null-token
+        // termination alone would refetch that token and fail as a
+        // PaginationLoop; the has-more signal must end the scan after one
+        // request.
+        let gateway = MockGateway::start(|req| {
+            if req.method == "GET" && req.path == "/v1/health" {
+                return MockResponse::ok("{}");
+            }
+            if req.method == "GET" && req.path.starts_with("/v1/actions/") {
+                return feishu_discovery(&req.path);
+            }
+            if req.method == "POST" && req.path == "/v1/actions/feishu.list_wiki_spaces" {
+                return MockResponse::ok(&envelope_ok(
+                    &json!({"items": [{"space_id": "700000038", "name": "s"}],
+                             "pageToken": "0||7000000000000000001", "hasMore": false})
+                    .to_string(),
+                ));
+            }
+            MockResponse::new(404, "{}")
+        })
+        .await;
+        let (gateway, ctx) =
+            setup_with_gateway(gateway, "SKARDI_TEST_OC_FEISHU_WIKI_HM", "wiki_spaces").await;
+
+        let batches = collect(&ctx, "SELECT space_id FROM saas.ws.wiki_spaces").await;
+        assert_eq!(batches.iter().map(RecordBatch::num_rows).sum::<usize>(), 1);
+        assert_eq!(
+            execute_inputs(&gateway).len(),
+            1,
+            "hasMore:false ended the scan; the non-empty token was never refetched"
+        );
     }
 
     #[tokio::test]

crates/skardi/src/sources/providers/open_connector/packs/feishu.yaml

@@ -9,11 +9,16 @@
 # (`pageSize` / `pageToken` / `chatId`; `additionalProperties: false`).
 # Design rationale lives in the module docs of packs/feishu.rs.
 #
-# DRAFT COLUMN STATUS: item schemas are declared LOOSE by the gateway
-# (empty properties + additionalProperties), so every column below rides
-# passthrough and none is protected by the fingerprint gate. Column sets
-# are drafted from Feishu's API reference and MUST be reconciled against
-# real rows (skill phase 4) before this pack is submission-ready.
+# COLUMN STATUS: item schemas are declared LOOSE by the gateway (empty
+# properties + additionalProperties), so every column below rides
+# passthrough and none is protected by the fingerprint gate — real rows
+# are the only column truth. tasks / wiki_spaces / wiki_nodes are
+# reconciled against a live workspace (2026-08-04): tasks carry
+# status/completed_at (no `completed` boolean), wiki spaces answer the
+# final page with a NON-empty pageToken and hasMore:false (the reason
+# has_more_path exists). chats / messages / chat_members column sets
+# remain drafted from Feishu's API reference pending the workspace app's
+# bot capability (Feishu 232025 gates the im APIs on it).
 kind: pack
 pack: feishu
 version: 1
@@ -29,6 +34,7 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 100
+      has_more_path: "$.hasMore"
     # ByCreateTimeAsc is the STABLE ordering. The API default
     # (ByActiveTimeDesc) reshuffles rows whenever a chat receives a
     # message mid-scan, which can skip or duplicate chats across cursor
@@ -55,6 +61,7 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 100
+      has_more_path: "$.hasMore"
     # containerIdType is pinned to `chat`: this table IS chat history
     # (thread listing is a different action, list_thread_messages).
     # ByCreateTimeAsc for the same cursor-stability reason as chats —
@@ -103,6 +110,7 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 100
+      has_more_path: "$.hasMore"
     resources:
       required: [chatId]
     columns:
@@ -121,23 +129,28 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 100
+      has_more_path: "$.hasMore"
     # `my_tasks` is the executor's default; pinned explicitly so the
-    # table's population is declared, not inherited. Omitting `completed`
-    # returns tasks in BOTH states — the state=all default.
+    # table's population is declared, not inherited. Omitting the
+    # action's `completed` input returns tasks in BOTH states — the
+    # state=all default. The input is deliberately unmapped as a filter:
+    # real rows carry NO `completed` boolean (live-verified 2026-08-04 —
+    # completion is `status: todo|done` plus `completed_at`), and a
+    # string-enum column cannot render into a boolean input without a
+    # value transform the engine deliberately lacks.
     fixed_inputs:
       type: my_tasks
-    filters:
-      # Boolean equality forwards faithfully, but stays Inexact like
-      # every enum-ish push — re-filtering is cheap and the provider's
-      # interpretation is not contract-tested.
-      - { column: completed, op: eq, input: completed, fidelity: inexact }
     columns:
       - { name: guid, path: guid, type: utf8, nullable: false }
       - { name: summary, path: summary, type: utf8, nullable: true }
       - { name: description, path: description, type: utf8, nullable: true }
-      - { name: completed, path: completed, type: boolean, nullable: true }
+      - { name: status, path: status, type: utf8, nullable: true }
       - { name: created_at, path: created_at, type: timestamp_ms_string_utc, nullable: true }
       - { name: updated_at, path: updated_at, type: timestamp_ms_string_utc, nullable: true }
+      # Feishu spells "not completed" as the digit string "0", which
+      # decodes to 1970-01-01T00:00:00Z — filter on `status = 'done'`
+      # rather than on this column's epoch-zero sentinel.
+      - { name: completed_at, path: completed_at, type: timestamp_ms_string_utc, nullable: true }
       - { name: due, path: due, type: json, nullable: true }
       - { name: members, path: members, type: json, nullable: true }
       - { name: url, path: url, type: utf8, nullable: true }
@@ -152,12 +165,14 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 50
+      has_more_path: "$.hasMore"
     columns:
       - { name: space_id, path: space_id, type: utf8, nullable: false }
       - { name: name, path: name, type: utf8, nullable: true }
       - { name: description, path: description, type: utf8, nullable: true }
       - { name: space_type, path: space_type, type: utf8, nullable: true }
       - { name: visibility, path: visibility, type: utf8, nullable: true }
+      - { name: open_sharing, path: open_sharing, type: utf8, nullable: true }
 
   wiki_nodes:
     action: feishu.list_wiki_nodes
@@ -169,6 +184,7 @@ tables:
       next_cursor_path: "$.pageToken"
       page_size_input: pageSize
       page_size: 50
+      has_more_path: "$.hasMore"
     # One level per scan: the action lists the children of ONE parent
     # (space root when parentNodeToken is omitted) — same shape as
     # Notion's block_children. Full-tree traversal is a client-side
@@ -189,3 +205,5 @@ tables:
       - { name: obj_edit_time, path: obj_edit_time, type: timestamp_s_string_utc, nullable: true }
       - { name: node_create_time, path: node_create_time, type: timestamp_s_string_utc, nullable: true }
       - { name: owner, path: owner, type: utf8, nullable: true }
+      - { name: creator, path: creator, type: utf8, nullable: true }
+      - { name: url, path: url, type: utf8, nullable: true }