fix: restore the extractor entrypoint and regenerate the two artifacts

Three CI failures, one of them a real runtime break.

get_all_conversations had been deleted from LinkedInExtractor by an earlier
regex edit that was meant to restore get_inbox. The MCP tool calls that method,
so the tool would have failed at runtime. Nothing caught it: every test here
either drives the reader directly or mocks the extractor, so the passthrough
itself was untested. ty found it.

Added two tests that would have: one asserts the method exists, and one spies
on the reader and compares the forwarded keyword set against the reader's own
signature, so a silently dropped argument fails too.

Regenerated docs/scraping-architecture.md and the facade-contract policy trace,
both of which enumerate the scraping surface and so change when a tool is added.
The trace generator refuses to write into the canonical fixture directory, so it
was generated elsewhere and the single changed file copied in deliberately:
90 lines added, none removed.

All pre-commit hooks pass.

Author: tmedford

docs/scraping-architecture.md

@@ -40,6 +40,7 @@ a page-owning collaborator.
 | `search_urls` | `CONTENT_DATE_POSTED_MAP`, `EXPERIENCE_LEVEL_MAP`, `JOB_DATE_POSTED_MAP`, `JOB_TYPE_MAP`, `NETWORK_TOKENS`, `SORT_BY_MAP`, `WORK_TYPE_MAP`, `build_company_search_url()`, `build_content_search_url()`, `build_job_search_url()`, `build_people_search_url()` | `browser-free` |
 | `session` | `NAV_DELAY`, `ScrapingSession` | `page-owning` |
 | `text` | `DETAIL_CAPTURE_EN_US`, `DetailCaptureTextTable`, `SIDEBAR_CHROME_EN`, `SidebarChromeTable`, `filter_linkedin_noise_lines()`, `strip_conversation_chrome()`, `strip_linkedin_noise()`, `truncate_linkedin_noise()` | `browser-free` |
+| `voyager_messaging` | `KNOWN_CATEGORIES`, `MAX_PAGE_SIZE`, `MESSAGING_URL`, `VoyagerMessagingReader` | `page-owning` |
 
 ## Internal import graph
 
@@ -51,7 +52,7 @@ a page-owning collaborator.
 - `content` -> `session`, `text`
 - `contracts` -> `identifiers`, `link_metadata`
 - `conversations` -> `content`, `identifiers`, `link_metadata`, `navigation`, `profile_page`, `session`, `text`
-- `extractor` -> `capture`, `company`, `connection_actions`, `content`, `contracts`, `conversations`, `feed`, `job_pages`, `jobs`, `message_sender`, `navigation`, `person`, `posts`, `profile_page`, `session`, `text`
+- `extractor` -> `capture`, `company`, `connection_actions`, `content`, `contracts`, `conversations`, `feed`, `job_pages`, `jobs`, `message_sender`, `navigation`, `person`, `posts`, `profile_page`, `session`, `text`, `voyager_messaging`
 - `feed` -> `content`, `contracts`, `feed_payload`, `navigation`, `session`, `text`
 - `feed_payload` -> `link_metadata`
 - `fields` -> `capture`
@@ -68,13 +69,15 @@ a page-owning collaborator.
 - `search_urls` -> `contracts`
 - `session` -> _(none)_
 - `text` -> _(none)_
+- `voyager_messaging` -> _(none)_
 
 ## `LinkedInExtractor` public coroutine surface
 
 - `click_button_by_text`
 - `connect_with_person`
 - `extract_feed`
 - `extract_page`
+- `get_all_conversations`
 - `get_company_employees`
 - `get_conversation`
 - `get_inbox`
@@ -104,6 +107,7 @@ a page-owning collaborator.
 - `_message_sender`
 - `_person`
 - `_posts`
+- `_voyager_messaging`
 
 ## Dependency-direction violations
 

linkedin_mcp_server/scraping/extractor.py

@@ -230,6 +230,29 @@ async def get_inbox(self, limit: int = 20) -> dict[str, Any]:
         """List recent conversations from the messaging inbox."""
         return await self._conversations.get_inbox(limit)
 
+    async def get_all_conversations(
+        self,
+        limit: int = 200,
+        max_pages: int = 60,
+        cursor: str | None = None,
+        quiet_for_days: int | None = None,
+        awaiting_reply_only: bool = False,
+        category: str | None = None,
+        page_size: int = 25,
+        known_thread_urns: list[str] | None = None,
+    ) -> dict[str, Any]:
+        """Page the whole mailbox via the conversations API, clicking nothing."""
+        return await self._voyager_messaging.get_all_conversations(
+            limit=limit,
+            max_pages=max_pages,
+            cursor=cursor,
+            quiet_for_days=quiet_for_days,
+            awaiting_reply_only=awaiting_reply_only,
+            category=category,
+            page_size=page_size,
+            stop_at_thread_urns=set(known_thread_urns or ()) or None,
+        )
+
     async def get_conversation(
         self,
         linkedin_username: str | None = None,

tests/fixtures/scraping-policy/v1/facade-contract.json

@@ -156,6 +156,96 @@
           "type": "object"
         }
       },
+      "get_all_conversations": {
+        "input": {
+          "additionalProperties": false,
+          "properties": {
+            "awaiting_reply_only": {
+              "default": false,
+              "description": "only return threads whose newest message is\ntheirs, so a reply is owed",
+              "type": "boolean"
+            },
+            "category": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "default": null,
+              "description": "SERVER-SIDE filter, the only one LinkedIn actually honours.\nOne of INBOX, PRIMARY_INBOX, ARCHIVE, INMAIL, STARRED, SPAM.\nThese jump anywhere in time in a single call. An unknown value\nis rejected rather than passed through, because the API answers\none with an empty page rather than an error."
+            },
+            "cursor": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "default": null,
+              "description": "next_cursor from a previous call, to resume the walk"
+            },
+            "known_thread_urns": {
+              "anyOf": [
+                {
+                  "items": {
+                    "type": "string"
+                  },
+                  "type": "array"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "default": null,
+              "description": "thread urns already recorded elsewhere. The\nmailbox is recency-ordered, so once a whole page is threads you\nalready know, everything behind it is older and also known, and\nthe walk stops. This is what keeps a repeat sync to one or two\npages instead of the whole mailbox."
+            },
+            "limit": {
+              "default": 200,
+              "description": "Maximum conversations to return (1-2000, default 200)",
+              "maximum": 2000,
+              "minimum": 1,
+              "type": "integer"
+            },
+            "max_pages": {
+              "default": 60,
+              "description": "Safety cap on cursor pages to walk (1-200, default 60)",
+              "maximum": 200,
+              "minimum": 1,
+              "type": "integer"
+            },
+            "page_size": {
+              "default": 25,
+              "description": "rows per request, max 25 (measured). Above 25 the API\nreturns EMPTY rather than an error, so this is clamped.",
+              "maximum": 25,
+              "minimum": 1,
+              "type": "integer"
+            },
+            "quiet_for_days": {
+              "anyOf": [
+                {
+                  "minimum": 1,
+                  "type": "integer"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "default": null,
+              "description": "only return threads with no activity for this many\ndays — the reconnect filter"
+            }
+          },
+          "type": "object"
+        },
+        "output": {
+          "additionalProperties": true,
+          "type": "object"
+        }
+      },
       "get_company_employees": {
         "input": {
           "additionalProperties": false,

tests/test_voyager_messaging.py

@@ -406,3 +406,58 @@ async def test_filtered_walks_still_request_full_pages(self):
         reader = _Reader([_payload(rows, None)])
         await reader.get_all_conversations(limit=5, page_size=25, quiet_for_days=365)
         assert "count:25" in reader.fetched[0], "filtered walk keeps full pages"
+
+
+class TestExtractorWiring:
+    """The tool calls `extractor.get_all_conversations`. Every other test here
+    either drives the reader directly or mocks the extractor, so none of them
+    would notice the method going missing -- which is exactly what happened
+    during a refactor, caught only by the type checker."""
+
+    def test_extractor_exposes_the_tool_entrypoint(self):
+        from linkedin_mcp_server.scraping.extractor import LinkedInExtractor
+
+        assert hasattr(LinkedInExtractor, "get_all_conversations")
+
+    async def test_extractor_forwards_every_argument_to_the_reader(self):
+        """A passthrough that silently drops an argument is the other way this
+        breaks without any test noticing."""
+        import inspect
+
+        from linkedin_mcp_server.scraping.extractor import LinkedInExtractor
+
+        seen: dict = {}
+
+        class _Spy:
+            async def get_all_conversations(self, **kwargs):
+                seen.update(kwargs)
+                return {"conversations": [], "count": 0}
+
+        extractor = LinkedInExtractor.__new__(LinkedInExtractor)
+        extractor._voyager_messaging = _Spy()  # type: ignore[attr-defined]
+
+        await extractor.get_all_conversations(
+            limit=7,
+            max_pages=3,
+            cursor="CUR",
+            quiet_for_days=30,
+            awaiting_reply_only=True,
+            category="INMAIL",
+            page_size=11,
+            known_thread_urns=["a", "b"],
+        )
+
+        assert seen["limit"] == 7
+        assert seen["max_pages"] == 3
+        assert seen["cursor"] == "CUR"
+        assert seen["quiet_for_days"] == 30
+        assert seen["awaiting_reply_only"] is True
+        assert seen["category"] == "INMAIL"
+        assert seen["page_size"] == 11
+        assert seen["stop_at_thread_urns"] == {"a", "b"}
+
+        # Every reader parameter the extractor is meant to relay must be relayed.
+        reader_params = set(
+            inspect.signature(VoyagerMessagingReader.get_all_conversations).parameters
+        ) - {"self"}
+        assert reader_params == set(seen), reader_params ^ set(seen)