@consiliency/agent-board-schema 1.2.0

package/migrations/20260513000002_agent_board_monitoring_views.sql

@@ -0,0 +1,227 @@
+insert into public.app_settings(key, value)
+values
+  ('agent_board_monitoring_stale_claim_hours', to_jsonb(24)),
+  ('agent_board_monitoring_presence_lag_minutes', to_jsonb(15)),
+  ('agent_board_monitoring_callback_retry_threshold', to_jsonb(25)),
+  ('agent_board_monitoring_inbox_pending_threshold', to_jsonb(50)),
+  ('agent_board_monitoring_table_growth_bytes', to_jsonb(52428800))
+on conflict (key) do nothing;
+
+create or replace view agent_board.v_ops_stale_claims
+with (security_invoker = true)
+as
+select
+  c.id as claim_id,
+  c.task_id,
+  c.thread_id,
+  c.agent_id,
+  a.handle as agent_handle,
+  c.resource_kind,
+  c.resource_key,
+  round(extract(epoch from (now() - c.created_at)) / 3600.0, 2) as stale_hours,
+  c.lease_expires_at,
+  c.created_at
+from agent_board.claims c
+join agent_board.agents a on a.id = c.agent_id
+where c.status = 'active'
+  and c.released_at is null
+  and c.lease_expires_at < now();
+
+create or replace view agent_board.v_ops_presence_heartbeats
+with (security_invoker = true)
+as
+select
+  p.agent_id,
+  a.handle as agent_handle,
+  p.availability,
+  p.last_heartbeat_at,
+  p.expires_at,
+  round(extract(epoch from (now() - p.last_heartbeat_at)) / 60.0, 2) as heartbeat_lag_minutes,
+  (p.expires_at < now()) as is_expired
+from agent_board.presence p
+join agent_board.agents a on a.id = p.agent_id
+where p.expires_at < now();
+
+create or replace view agent_board.v_ops_open_escalations
+with (security_invoker = true)
+as
+select
+  e.id as escalation_id,
+  e.task_id,
+  e.thread_id,
+  e.severity,
+  e.status,
+  e.summary,
+  e.assigned_agent_id,
+  assignee.handle as assigned_agent_handle,
+  round(extract(epoch from (now() - e.created_at)) / 3600.0, 2) as age_hours,
+  e.created_at
+from agent_board.escalations e
+left join agent_board.agents assignee on assignee.id = e.assigned_agent_id
+where e.status in ('open', 'acknowledged');
+
+create or replace view agent_board.v_ops_callback_retry_pressure
+with (security_invoker = true)
+as
+select
+  coalesce(vcdh.endpoint_id, '00000000-0000-0000-0000-000000000000'::uuid) as endpoint_id,
+  count(*) filter (where vcdh.latest_attempt_outcome = 'retryable_failure')::bigint as retryable_failure_count,
+  count(*) filter (where vcdh.latest_outbox_status = 'retryable')::bigint as queued_retry_count,
+  max(vcdh.latest_attempt_finished_at) as latest_retry_finished_at,
+  max(vcdh.last_failure_at) as endpoint_last_failure_at
+from agent_board.v_callback_delivery_health vcdh
+group by coalesce(vcdh.endpoint_id, '00000000-0000-0000-0000-000000000000'::uuid)
+having count(*) filter (where vcdh.latest_attempt_outcome = 'retryable_failure') > 0
+    or count(*) filter (where vcdh.latest_outbox_status = 'retryable') > 0;
+
+create or replace view agent_board.v_ops_endpoint_disablement
+with (security_invoker = true)
+as
+select
+  ce.id as endpoint_id,
+  ce.owner_agent_id,
+  owner.handle as owner_agent_handle,
+  ce.disabled_reason,
+  ce.disabled_at,
+  ce.last_failure_at
+from agent_board.callback_endpoints ce
+join agent_board.agents owner on owner.id = ce.owner_agent_id
+where ce.endpoint_status = 'disabled';
+
+create or replace view agent_board.v_ops_inbox_backlog
+with (security_invoker = true)
+as
+select
+  counts.recipient_agent_id as agent_id,
+  a.handle as recipient_handle,
+  counts.unread_count,
+  counts.pending_action_count,
+  counts.total_visible_count,
+  counts.latest_inbox_updated_at
+from agent_board.v_agent_inbox_counts counts
+join agent_board.agents a on a.id = counts.recipient_agent_id
+where counts.pending_action_count > 0
+   or counts.unread_count > 0;
+
+create or replace view agent_board.v_ops_table_growth
+with (security_invoker = true)
+as
+select
+  table_name,
+  pg_total_relation_size(format('agent_board.%I', table_name)::regclass) as total_bytes
+from (
+  values
+    ('messages'),
+    ('task_events'),
+    ('presence'),
+    ('callback_outbox_events'),
+    ('callback_delivery_attempts'),
+    ('inbox_items'),
+    ('delivery_recipients'),
+    ('board_memberships')
+) as monitored(table_name);
+
+create or replace view agent_board.v_callback_health
+with (security_invoker = true)
+as
+with subscription_counts as (
+  select
+    cs.endpoint_id,
+    count(*) filter (where cs.subscription_status = 'active')::bigint as active_subscription_count,
+    count(*) filter (where cs.subscription_status = 'disabled')::bigint as disabled_subscription_count
+  from agent_board.callback_subscriptions cs
+  group by cs.endpoint_id
+),
+attempt_counts as (
+  select
+    cda.endpoint_id,
+    count(*) filter (where cda.attempt_outcome = 'delivered')::bigint as delivered_count,
+    count(*) filter (where cda.attempt_outcome = 'retryable_failure')::bigint as retryable_failure_count,
+    count(*) filter (where cda.attempt_outcome = 'terminal_failure')::bigint as terminal_failure_count,
+    count(*) filter (where cda.attempt_outcome = 'endpoint_disabled')::bigint as endpoint_disabled_count,
+    max(cda.finished_at) as latest_attempt_finished_at
+  from agent_board.callback_delivery_attempts cda
+  group by cda.endpoint_id
+),
+outbox_counts as (
+  select
+    coe.endpoint_id,
+    count(*) filter (where coe.outbox_status = 'retryable')::bigint as queued_retry_count,
+    count(*) filter (where coe.outbox_status = 'disabled')::bigint as disabled_outbox_count,
+    min(coe.available_at) filter (where coe.outbox_status = 'retryable') as oldest_retry_available_at
+  from agent_board.callback_outbox_events coe
+  group by coe.endpoint_id
+),
+latest_outbox as (
+  select distinct on (coe.endpoint_id)
+    coe.endpoint_id,
+    coe.id as latest_outbox_event_id,
+    coe.outbox_status as latest_outbox_status,
+    coe.delivery_mode as latest_delivery_mode,
+    coe.source_kind as latest_source_kind,
+    coe.updated_at as latest_outbox_updated_at
+  from agent_board.callback_outbox_events coe
+  order by coe.endpoint_id, coe.updated_at desc, coe.created_at desc, coe.id desc
+)
+select
+  ce.id as endpoint_id,
+  ce.owner_agent_id,
+  owner.handle as owner_agent_handle,
+  ce.endpoint_status,
+  ce.signing_key_version,
+  ce.disabled_at,
+  ce.disabled_reason,
+  coalesce(subscription_counts.active_subscription_count, 0::bigint) as active_subscription_count,
+  coalesce(subscription_counts.disabled_subscription_count, 0::bigint) as disabled_subscription_count,
+  coalesce(attempt_counts.delivered_count, 0::bigint) as delivered_count,
+  coalesce(attempt_counts.retryable_failure_count, 0::bigint) as retryable_failure_count,
+  coalesce(attempt_counts.terminal_failure_count, 0::bigint) as terminal_failure_count,
+  coalesce(attempt_counts.endpoint_disabled_count, 0::bigint) as endpoint_disabled_count,
+  coalesce(outbox_counts.queued_retry_count, 0::bigint) as queued_retry_count,
+  coalesce(outbox_counts.disabled_outbox_count, 0::bigint) as disabled_outbox_count,
+  latest_outbox.latest_outbox_event_id,
+  latest_outbox.latest_outbox_status,
+  latest_outbox.latest_delivery_mode,
+  latest_outbox.latest_source_kind,
+  latest_attempt.attempt_number as latest_attempt_number,
+  latest_attempt.attempt_outcome as latest_attempt_outcome,
+  latest_attempt.http_status as latest_http_status,
+  latest_attempt.error_summary as latest_error_summary,
+  latest_attempt.next_retry_at as latest_next_retry_at,
+  attempt_counts.latest_attempt_finished_at,
+  latest_outbox.latest_outbox_updated_at,
+  outbox_counts.oldest_retry_available_at,
+  ce.last_delivery_at,
+  ce.last_failure_at,
+  round(extract(epoch from (now() - ce.last_delivery_at)))::bigint as last_delivery_age_seconds,
+  round(extract(epoch from (now() - ce.last_failure_at)))::bigint as last_failure_age_seconds,
+  round(extract(epoch from (now() - attempt_counts.latest_attempt_finished_at)))::bigint as latest_attempt_age_seconds,
+  round(extract(epoch from (now() - outbox_counts.oldest_retry_available_at)))::bigint as oldest_retry_age_seconds
+from agent_board.callback_endpoints ce
+join agent_board.agents owner on owner.id = ce.owner_agent_id
+left join subscription_counts on subscription_counts.endpoint_id = ce.id
+left join attempt_counts on attempt_counts.endpoint_id = ce.id
+left join outbox_counts on outbox_counts.endpoint_id = ce.id
+left join latest_outbox on latest_outbox.endpoint_id = ce.id
+left join lateral (
+  select
+    cda.attempt_number,
+    cda.attempt_outcome,
+    cda.http_status,
+    cda.error_summary,
+    cda.next_retry_at
+  from agent_board.callback_delivery_attempts cda
+  where cda.outbox_event_id = latest_outbox.latest_outbox_event_id
+  order by cda.attempt_number desc
+  limit 1
+) latest_attempt on true;
+
+grant select on agent_board.v_ops_stale_claims,
+  agent_board.v_ops_presence_heartbeats,
+  agent_board.v_ops_open_escalations,
+  agent_board.v_ops_callback_retry_pressure,
+  agent_board.v_ops_endpoint_disablement,
+  agent_board.v_ops_inbox_backlog,
+  agent_board.v_ops_table_growth,
+  agent_board.v_callback_health
+to authenticated, service_role;

package/migrations/20260514000000_agent_board_auth_rls_id_qualification.sql

@@ -0,0 +1,150 @@
+-- Fix ambiguous column resolution in agent_board RLS-policy EXISTS subqueries
+-- introduced by 20260512000003_agent_board_auth_rls.sql.
+--
+-- Bug class (deny): a subquery's FROM alias introduces a column (`id`) that
+-- shadows the outer table's `id`. PostgreSQL resolves the unqualified column
+-- to the inner alias, collapsing the join predicate to a tautological
+-- comparison that is essentially never true. Effect: legitimate readers
+-- (joined board members, board-delivery recipients) are denied access.
+--   - agent_board_authenticated_select_messages (line 136):
+--       dr.source_message_id = id  -> dr.source_message_id = dr.id
+--   - agent_board_authenticated_select_tasks (line 165):
+--       dr.source_task_id = id     -> dr.source_task_id = dr.id
+--   - agent_board_authenticated_select_boards (line 354):
+--       bm.board_id = id           -> bm.board_id = bm.id
+--
+-- Bug class (leak): a subquery's FROM alias introduces a column whose name
+-- matches the outer table's column (`board_id` on board_memberships and on
+-- delivery_recipients/inbox_items). The unqualified outer reference resolves
+-- to the inner alias instead, collapsing the predicate to bm.board_id =
+-- bm.board_id (always true for any non-null membership row). Effect: an
+-- authenticated agent who is a joined/subscribed member of ANY board can
+-- read delivery_recipients and inbox_items rows scoped to ANY OTHER board.
+--   - agent_board_authenticated_select_delivery_recipients (line 383):
+--       bm.board_id = board_id  -> bm.board_id = bm.board_id
+--   - agent_board_authenticated_select_inbox_items (line 404):
+--       bm.board_id = board_id  -> bm.board_id = bm.board_id
+--
+-- Fix: drop and recreate the affected policies, qualifying every outer-table
+-- reference (messages.id, tasks.id, boards.id, delivery_recipients.board_id,
+-- inbox_items.board_id). No schema-shape change; this is a behavior fix to
+-- the policy expressions.
+
+drop policy if exists agent_board_authenticated_select_messages on agent_board.messages;
+create policy agent_board_authenticated_select_messages
+on agent_board.messages
+as permissive
+for select
+to authenticated
+using (
+  author_agent_id = agent_board.current_agent_id()
+  or exists (
+    select 1
+      from agent_board.delivery_recipients dr
+     where dr.source_message_id = messages.id
+       and (
+         dr.recipient_agent_id = agent_board.current_agent_id()
+         or (
+           dr.board_id is not null
+           and exists (
+             select 1
+               from agent_board.board_memberships bm
+              where bm.board_id = dr.board_id
+                and bm.agent_id = agent_board.current_agent_id()
+                and bm.membership_state in ('joined', 'subscribed')
+           )
+         )
+       )
+  )
+);
+
+drop policy if exists agent_board_authenticated_select_tasks on agent_board.tasks;
+create policy agent_board_authenticated_select_tasks
+on agent_board.tasks
+as permissive
+for select
+to authenticated
+using (
+  created_by_agent_id = agent_board.current_agent_id()
+  or assigned_agent_id = agent_board.current_agent_id()
+  or exists (
+    select 1
+      from agent_board.delivery_recipients dr
+     where dr.source_task_id = tasks.id
+       and (
+         dr.recipient_agent_id = agent_board.current_agent_id()
+         or (
+           dr.board_id is not null
+           and exists (
+             select 1
+               from agent_board.board_memberships bm
+              where bm.board_id = dr.board_id
+                and bm.agent_id = agent_board.current_agent_id()
+                and bm.membership_state in ('joined', 'subscribed')
+           )
+         )
+       )
+  )
+);
+
+drop policy if exists agent_board_authenticated_select_boards on agent_board.boards;
+create policy agent_board_authenticated_select_boards
+on agent_board.boards
+as permissive
+for select
+to authenticated
+using (
+  created_by_agent_id = agent_board.current_agent_id()
+  or exists (
+    select 1
+      from agent_board.board_memberships bm
+     where bm.board_id = boards.id
+       and bm.agent_id = agent_board.current_agent_id()
+       and bm.membership_state in ('joined', 'subscribed')
+  )
+);
+
+drop policy if exists agent_board_authenticated_select_delivery_recipients on agent_board.delivery_recipients;
+create policy agent_board_authenticated_select_delivery_recipients
+on agent_board.delivery_recipients
+as permissive
+for select
+to authenticated
+using (
+  recipient_agent_id = agent_board.current_agent_id()
+  or (
+    board_id is not null
+    and exists (
+      select 1
+        from agent_board.board_memberships bm
+       where bm.board_id = delivery_recipients.board_id
+         and bm.agent_id = agent_board.current_agent_id()
+         and bm.membership_state in ('joined', 'subscribed')
+    )
+  )
+);
+
+drop policy if exists agent_board_authenticated_select_inbox_items on agent_board.inbox_items;
+create policy agent_board_authenticated_select_inbox_items
+on agent_board.inbox_items
+as permissive
+for select
+to authenticated
+using (
+  recipient_agent_id = agent_board.current_agent_id()
+  or (
+    board_id is not null
+    and visible_in_board_feed = true
+    and exists (
+      select 1
+        from agent_board.board_memberships bm
+       where bm.board_id = inbox_items.board_id
+         and bm.agent_id = agent_board.current_agent_id()
+         and bm.membership_state in ('joined', 'subscribed')
+    )
+  )
+);
+
+insert into agent_board.schema_version (version)
+values (4)
+on conflict ((true)) do update set version = excluded.version;