From bc1ee9d4002b3112a9a71ff717dd452fafe6b29a Mon Sep 17 00:00:00 2001
From: Rodrigo Gomes da Silva <rodrigo.smscom@gmail.com>
Date: Fri, 14 Aug 2026 20:43:39 -0300
Subject: [PATCH] fix(sessions): run the assigned recovery pass in the
 all-profiles view
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

`?all_profiles=1` passed `project_assigned_limit=False`, on the premise that
its `visible_session_limit=None` first pass already returns every conversation.
It does not: `_load_cli_sessions_uncached` maps `visible_session_limit is None`
to CLI_VISIBLE_SESSION_LIMIT for the interactive pass, so that view truncates
by recency exactly like the single-profile one. Only cron/webhook/kanban limits
reach the reader as `limit=` directly, where None does mean unbounded.

Net effect of the old value: in the "All profiles" sidebar a project-assigned
CLI session older than the newest 20 interactive conversations was missing from
/api/sessions while its project chip still claimed it — the #6659 bug itself,
in the one view that shows every profile at once. Recovery pass 2 (the
unassigned refill) still ran there, so the two passes disagreed as well.

Pass PROJECT_ASSIGNED_CLI_LIMIT instead of None so the per-project bound and
the PROJECT_ASSIGNED_CLI_SCAN_CEILING guard apply per profile, rather than
issuing an unbounded assigned scan for every profile on the box.
---
 api/models.py                                 | 13 ++++--
 ...test_project_assigned_cli_session_limit.py | 43 +++++++++++++++++++
 2 files changed, 52 insertions(+), 4 deletions(-)

diff --git a/api/models.py b/api/models.py
index 56265252..2cc0d857 100644
--- a/api/models.py
+++ b/api/models.py
@@ -8905,12 +8905,17 @@ def get_cli_sessions(
             merged: list[dict] = []
             for idx, (ctx_home, ctx_db_path, ctx_profile) in enumerate(contexts):
                 load_kwargs = {
+                    # NOTE: visible_session_limit=None is NOT "unbounded" for the
+                    # interactive pass — it resolves to CLI_VISIBLE_SESSION_LIMIT
+                    # above, so this view truncates assigned conversations by
+                    # recency exactly like the single-profile one and needs the
+                    # same recovery passes. project_assigned_limit therefore keeps
+                    # its default per-project bound here. Only the three limits
+                    # below are handed straight to the reader as ``limit=``, where
+                    # None really does mean unbounded.
                     'source_filter': source_filter,
                     'visible_session_limit': None,
-                    # An unbounded first pass already returns every assigned and
-                    # unassigned conversation, so the bounded recovery passes
-                    # have nothing to recover — skip their queries.
-                    'project_assigned_limit': False,
+                    'project_assigned_limit': PROJECT_ASSIGNED_CLI_LIMIT,
                     'cron_project_limit': None,
                     'webhook_project_limit': None,
                     'kanban_project_limit': None,
diff --git a/tests/test_project_assigned_cli_session_limit.py b/tests/test_project_assigned_cli_session_limit.py
index ca8829f6..d4d8ff31 100755
--- a/tests/test_project_assigned_cli_session_limit.py
+++ b/tests/test_project_assigned_cli_session_limit.py
@@ -738,3 +738,46 @@ def test_kanban_second_pass_still_projects_rows(fake_hermes_home, tmp_path):
     assert "kanban-old" in by_id
     # Upstream semantics: kanban rows get no chip from this helper.
     assert by_id["kanban-old"]["project_id"] is None
+
+
+# ── The all-profiles sidebar view runs the same recovery passes ───────────────
+
+
+def test_all_profiles_view_recovers_project_assigned_sessions(
+    fake_hermes_home, tmp_path, monkeypatch
+):
+    """``?all_profiles=1`` truncates by recency too, so it needs both passes.
+
+    That path passes ``visible_session_limit=None``, which reads like
+    "unbounded" but is mapped to ``CLI_VISIBLE_SESSION_LIMIT`` for the
+    interactive pass — only the cron/webhook/kanban limits reach the reader as
+    ``limit=`` directly, where None really does mean unbounded. Skipping the
+    assigned recovery pass here therefore lost exactly the sessions this
+    regression is about, in the one view that shows every profile at once.
+    """
+    monkeypatch.setattr(models, "CLI_VISIBLE_SESSION_LIMIT", 5)
+    _register_projects(tmp_path, "project-123")
+
+    db_path = fake_hermes_home / "state.db"
+    rows = [
+        _session(f"recent-{index:02d}", BASE_TS + 100 + index)
+        for index in range(25)
+    ]
+    rows.append(_session("older-assigned", BASE_TS, project_id="project-123"))
+    _write_state_db(db_path, rows)
+
+    # Enumerating real profiles shells out to the agent CLI; pin the one context.
+    monkeypatch.setattr(
+        models,
+        "_all_profiles_cli_contexts",
+        lambda: ([(fake_hermes_home, db_path, "default")], (("home", "default", 1),)),
+    )
+
+    sessions = models.get_cli_sessions(all_profiles=True)
+    by_id = {session["session_id"]: session for session in sessions}
+
+    assert "older-assigned" in by_id
+    assert by_id["older-assigned"]["project_id"] == "project-123"
+    assert [s["session_id"] for s in sessions].count("older-assigned") == 1
+    # The unassigned window is still bounded — the recovery pass is additive.
+    assert sum(1 for s in sessions if s["project_id"] is None) == 5