fix(zendesk): unassign tickets whose live chat is stuck open

docs/jobs/zendesk-digest.md

@@ -85,6 +85,10 @@ Two limits bound a message and whichever binds first splits it — 40 components
 
 **That is what lets everything here go over one incoming webhook** — the digest, the positive-review tally and the failure alerts alike. A webhook no application owns may send non-interactive components and nothing else, so adding an interactive one means moving the digest back to a bot token and a channel id. The request needs `?with_components=true` (`digest_webhook_url`): Discord ignores the `components` field on a webhook post without it, and the digest is nothing but components.
 
+## Live chats stuck open
+
+Zendesk writes a Twitter DM chat onto the ticket only when the chat ends, and it never ends a chat that started while the ticket had an assignee. The auto-assign trigger gives every ticket one, so a customer writing back on a solved DM ticket is invisible to everyone but the assignee. Each run sweeps the unsolved DM tickets for an open chat and unassigns them; the chat then ends about 10 minutes later and the message lands on the ticket. The digest lists them under 💬 and records none of them, so each is reported again with its newest message. A ticket Zendesk refused to unassign is listed apart, for an admin. `--dry-run`, `--no-discord` and a rehearsal list them without unassigning.
+
 ## Drafting the known answers
 
 Before posting, the digest drafts a private reply note on each ticket it reports that

src/session_ops/zendesk/api.py

@@ -234,6 +234,21 @@ def has_open_chat(session, subdomain, ticket_id):
     return False
 
 
+def unassign(session, subdomain, ticket_id):
+    """Clear the ticket's assignee, which is what lets Zendesk end its live chat.
+    Returns whether Zendesk took it."""
+    url = f"https://{subdomain}.zendesk.com/api/v2/tickets/{ticket_id}.json"
+    try:
+        resp = session.request("PUT", url, json={"ticket": {"assignee_id": None}})
+    except requests.RequestException as exc:
+        print(f"Note: could not unassign #{ticket_id} ({exc}).")
+        return False
+    if resp.status_code >= 400:
+        print(f"Note: unassigning #{ticket_id} returned {resp.status_code}.")
+        return False
+    return True
+
+
 STAR_SUBJECT = re.compile(r"^\s*([★☆]{1,10})")
 
 

src/session_ops/zendesk/triage.py

@@ -74,7 +74,7 @@
 from session_ops.shared.env import get_env, rehearsing
 from session_ops.shared.text import squash
 from session_ops.zendesk.api import (REVIEW_CHANNEL, SEARCH_RESULT_LIMIT, fetch_comments,
-                                     fetch_tickets, has_open_chat,
+                                     fetch_tickets, has_open_chat, unassign,
                                      hydrate_requester_activity, is_store_review,
                                      review_platform, review_stars, ticket_url,
                                      zendesk_session)
@@ -518,6 +518,16 @@ def find_open_chats(session, subdomain):
     return {t["id"] for t in tickets if has_open_chat(session, subdomain, t["id"])}
 
 
+def release_open_chats(session, subdomain, ticket_ids):
+    """Unassign each ticket so Zendesk ends its chat and writes the customer's newest
+    message onto it, about 10 minutes later. Returns the ids Zendesk accepted.
+
+    Zendesk never ends a chat that started while the ticket had an assignee, and the
+    auto-assign trigger gives every ticket one.
+    """
+    return {tid for tid in ticket_ids if unassign(session, subdomain, tid)}
+
+
 # ---- The classification request ---------------------------------------------
 
 
@@ -838,21 +848,34 @@ def build_non_english_line(findings, subdomain, updated_ids):
             f"{', '.join(links)}", set())
 
 
-def build_open_chat_line(ticket_ids, subdomain):
+def build_open_chat_line(ticket_ids, subdomain, released=()):
     """The tickets whose live chat is still open, so nobody but the assignee can read
     the customer's newest message. None when there are none.
 
-        💬 **2** live chats still open — newest message not on the ticket until an
-        admin unassigns it: #28835, #26946
+        💬 **2** live chats still open: unassigned, so the newest message reaches the
+        ticket in about 10 minutes: #28835, #26946
 
-    Covers no ticket: one stays listed every run until its chat ends.
+    A ticket the run could not unassign is listed apart, for an admin. Covers no
+    ticket: the conversation it was classified from is missing its newest message.
     """
     if not ticket_ids:
         return None
-    links = ", ".join(f"[#{tid}]({ticket_url(subdomain, tid)})" for tid in sorted(ticket_ids))
-    plural = "" if len(ticket_ids) == 1 else "s"
-    return (f"{OPEN_CHAT_MARKER} **{len(ticket_ids)}** live chat{plural} still open — "
-            f"newest message not on the ticket until an admin unassigns it: {links}", set())
+
+    def part(ids, outcome):
+        links = ", ".join(f"[#{tid}]({ticket_url(subdomain, tid)})" for tid in sorted(ids))
+        plural = "" if len(ids) == 1 else "s"
+        return f"{OPEN_CHAT_MARKER} **{len(ids)}** live chat{plural} still open: {outcome}: {links}"
+
+    released = set(released) & set(ticket_ids)
+    stuck = set(ticket_ids) - released
+    parts = []
+    if released:
+        parts.append(part(released, "unassigned, so the newest message reaches the ticket "
+                                    "in about 10 minutes"))
+    if stuck:
+        parts.append(part(stuck, "newest message not on the ticket until an admin "
+                                 "unassigns it"))
+    return "\n".join(parts), set()
 
 
 def build_drafted_line(ticket_ids, subdomain):
@@ -992,7 +1015,8 @@ def build_messages(findings, subdomain, stats=None, updated_ids=None):
     if omitted:
         header += (f"\nShowing the top **{len(shown)}** of "
                    f"**{len(shown) + len(omitted)}** worth looking into.")
-    open_chats = build_open_chat_line((stats or {}).get("open_chats"), subdomain)
+    open_chats = build_open_chat_line((stats or {}).get("open_chats"), subdomain,
+                                      (stats or {}).get("released_chats") or ())
     non_english = build_non_english_line(findings, subdomain, updated_ids)
     drafted = build_drafted_line((stats or {}).get("drafted"), subdomain)
     # First, under the header: the run's to-do lists, for an admin and for the `claude:` notes.
@@ -1126,6 +1150,11 @@ def main(argv=None):
         stats["open_chats"] = find_open_chats(zd, subdomain)
         if stats["open_chats"]:
             print(f"{len(stats['open_chats'])} ticket(s) have a live chat still open.")
+            # A counts-only run must not write to tickets.
+            if not (args.dry_run or args.no_discord or rehearsing()):
+                stats["released_chats"] = release_open_chats(zd, subdomain,
+                                                             stats["open_chats"])
+                print(f"Unassigned {len(stats['released_chats'])} of them.")
 
         # Drop positive store reviews before anything expensive: they were 59% of all
         # tickets in the sample and never actionable.

tests/zendesk/test_triage.py

@@ -794,8 +794,30 @@ def test_it_lists_the_swept_tickets_first(self):
         messages, _ = triage.build_messages(findings, "acme", {"open_chats": {28835, 26946}})
         lines = [l for l in digest_text(messages).splitlines() if "[#" in l]
         self.assertIn("**2** live chats still open", lines[0])
+        self.assertIn("until an admin unassigns it", lines[0])
         self.assertLess(lines[0].index("[#26946]"), lines[0].index("[#28835]"))
 
+    def test_released_tickets_are_listed_apart_from_stuck_ones(self):
+        stats = {"open_chats": {1, 2, 3}, "released_chats": {1, 3}}
+        messages, _ = triage.build_messages([finding(9)], "acme", stats)
+        lines = [l for l in digest_text(messages).splitlines() if "live chat" in l]
+        self.assertEqual(len(lines), 2)
+        self.assertIn("**2** live chats still open: unassigned", lines[0])
+        self.assertIn("[#1]", lines[0])
+        self.assertIn("[#3]", lines[0])
+        self.assertIn("**1** live chat still open: newest message not on the ticket", lines[1])
+        self.assertIn("[#2]", lines[1])
+
+    def test_release_unassigns_each_and_keeps_what_zendesk_took(self):
+        session = FakeSession([FakeResponse({}), FakeResponse({}, status_code=403)])
+        with contextlib.redirect_stdout(io.StringIO()):
+            released = triage.release_open_chats(session, "acme", [4, 5])
+        self.assertEqual(released, {4})
+        method, url, kwargs = session.calls[0]
+        self.assertEqual(method, "PUT")
+        self.assertTrue(url.endswith("/tickets/4.json"))
+        self.assertEqual(kwargs["json"], {"ticket": {"assignee_id": None}})
+
     def test_it_covers_no_ticket(self):
         _, coverage = triage.build_messages([finding(1)], "acme", {"open_chats": {9}})
         self.assertNotIn(9, set().union(*coverage))