feat(fusion_helpdesk_central): findings-first wizard, explicit Generate button

The old flow fired OpenAI on wizard open with just ticket + chatter,
so the AI summary was just a paraphrase of what the user originally
reported — your engineering analysis (scope, limitations, recommended
approach) never made it to the owner. Restructure to a two-step flow:

  1. Open wizard → empty findings + empty summary, NO OpenAI call
  2. You write findings: scope / effort / approach / risk
  3. Click 'Generate Summary from Findings' → OpenAI runs with
     ticket + chatter + findings, where the prompt explicitly tells
     the model to weight findings MORE THAN the original report
  4. Review/edit, then Send

Bulk wizard mirrors the flow per line: each row gets its own
findings + summary, one 'Generate All Summaries' button fans out
parallel OpenAI calls using each line's own findings.

Updated SUMMARY_PROMPT to:
- Tell the model the support engineer's findings are authoritative
- Emit a bullet structure that leads with the recommendation, not
  the user's restated ask
- Side with findings over the original report when they conflict

New tests cover:
- default_get does NOT fire OpenAI (regression guard for auto-AI)
- Findings text actually reaches the OpenAI prompt
- Send works with a manually-typed summary (no AI in the loop)
- Existing bulk + validation paths still pass with the new shape

Also folds in the deferred code-review #7: ThreadPoolExecutor now
explicitly cancels pending futures on timeout via
shutdown(wait=False, cancel_futures=True) so a slow OpenAI day can't
hold the wizard open for ceil(N/workers)*15s.

Bumps fusion_helpdesk_central to 19.0.2.3.0.

Smoke-tested live on nexa: opening the wizard makes zero OpenAI calls;
clicking Generate with findings='My findings: scope is XL, ~8h' makes
exactly one call and the findings text is verifiably in the prompt
body received by call_openai_chat.

docs/superpowers/specs/2026-05-27-nexacloud-billing-importer-design.md

@@ -183,6 +183,13 @@ until this is granted.
 - Importing historical NexaCloud invoices / `usage_records` (2d reads NexaCloud actuals).
 - Add-ons (`deployment_addons`) as recurring lines — revisit if material.
 
+> **Flip-day note (carry into 2b):** the inbound `/usage` API resolves a subscription by
+> its **Odoo integer id** (`int(subscription_external_id)`), but imported shadow subs are
+> keyed by NexaCloud's UUID in `x_fc_nexacloud_subscription_id`. Before NexaCloud can push
+> usage (2b), decide how it learns the Odoo id (return the mapping from the importer, or
+> extend the usage API to also resolve by `x_fc_nexacloud_subscription_id`). Not a 2a bug
+> (2a is read-only), but it must be resolved before the flip.
+
 ## 11. Verify at implementation (do NOT code from memory — CLAUDE rule #1)
 
 Confirm on odoo-trial Enterprise before relying on them:

fusion_centralize_billing/models/account_link.py

@@ -46,7 +46,9 @@ class FusionBillingAccountLink(models.Model):
             return existing
         partner = self.env['res.partner']
         if email:
-            partner = partner.search([('email', '=', email)], limit=1)
+            # case-insensitive so a pre-existing partner with a differently-cased email
+            # (created via the web UI or another sync) is reused, not duplicated.
+            partner = partner.search([('email', '=ilike', email)], limit=1)
         if not partner:
             partner = partner.create({'name': name or external_id, 'email': email, **(extra or {})})
         return self.create({

fusion_centralize_billing/tests/test_importer.py

@@ -211,6 +211,29 @@ class TestImporterErrorIsolation(TransactionCase):
         self.assertTrue(summary['failed'], "the bad row must be recorded in failed[]")
         self.assertTrue(any(f['kind'] == 'user' for f in summary['failed']))
 
+    def test_unknown_billing_cycle_is_failed_not_silently_monthly(self):
+        data = _fixture()
+        data['subscriptions'][0]['billing_cycle'] = 'annual'  # not monthly/yearly
+        summary = self.Wizard._import_rows(data)
+        self.assertFalse(self.env['sale.order'].search(
+            [('x_fc_nexacloud_subscription_id', '=', 's-1')]),
+            "an unrecognized billing_cycle must NOT silently create a monthly sub")
+        self.assertTrue(any(f['kind'] == 'subscription' and f['id'] == 's-1'
+                            for f in summary['failed']))
+
+    def test_missing_price_for_cycle_is_failed_not_zero(self):
+        data = _fixture()
+        data['plans'][0]['price_yearly'] = None   # s-2 is yearly -> no price for it
+        summary = self.Wizard._import_rows(data)
+        # the yearly sub fails (no silent $0 line); the monthly one still imports
+        self.assertFalse(self.env['sale.order'].search(
+            [('x_fc_nexacloud_subscription_id', '=', 's-2')]),
+            "a missing price for the cycle must NOT silently create a $0 line")
+        self.assertTrue(self.env['sale.order'].search(
+            [('x_fc_nexacloud_subscription_id', '=', 's-1')]))
+        self.assertTrue(any(f['kind'] == 'subscription' and f['id'] == 's-2'
+                            for f in summary['failed']))
+
 
 @tagged('post_install', '-at_install')
 class TestImporterReadGuard(TransactionCase):

fusion_centralize_billing/views/import_wizard_views.xml

@@ -5,6 +5,13 @@
         <field name="model">fusion.billing.import.wizard</field>
         <field name="arch" type="xml">
             <form string="Import from NexaCloud">
+                <div class="alert alert-danger" role="alert" invisible="failed_count == 0">
+                    <strong>Import completed with errors: </strong>
+                    <field name="failed_count" class="oe_inline" readonly="1"/> row(s) failed — see Result below.
+                </div>
+                <div class="alert alert-warning" role="alert" invisible="skipped_count == 0">
+                    <field name="skipped_count" class="oe_inline" readonly="1"/> row(s) skipped (unresolved customer/plan) — see Result below.
+                </div>
                 <group>
                     <field name="dry_run"/>
                 </group>

fusion_centralize_billing/wizards/import_wizard.py

@@ -33,12 +33,25 @@ class FusionBillingImportWizard(models.TransientModel):
         default=True,
         help="Read and report what would be imported, without writing anything.")
     result_summary = fields.Text(readonly=True)
+    failed_count = fields.Integer(readonly=True)
+    skipped_count = fields.Integer(readonly=True)
 
     def action_run_import(self):
         self.ensure_one()
         data = self._read_nexacloud_rows()
         summary = self._import_rows(data, dry_run=self.dry_run)
+        failed = summary.get("failed") or []
+        skipped = summary.get("skipped") or []
         self.result_summary = json.dumps(summary, indent=2, default=str)
+        self.failed_count = len(failed)
+        self.skipped_count = len(skipped)
+        # A partial billing import must be loud, not buried in the JSON. Log at ERROR
+        # so it survives nexa's log_level=warn (INFO is suppressed there).
+        if failed:
+            _logger.error("NexaCloud import: %s row(s) FAILED%s: %s",
+                          len(failed), " (dry-run)" if self.dry_run else "", failed)
+        if skipped:
+            _logger.warning("NexaCloud import: %s row(s) skipped: %s", len(skipped), skipped)
         return {
             "type": "ir.actions.act_window",
             "res_model": self._name,
@@ -83,6 +96,13 @@ class FusionBillingImportWizard(models.TransientModel):
                 "current_period_start, current_period_end FROM subscriptions")
             data["subscriptions"] = [dict(r) for r in cur.fetchall()]
             return data
+        except psycopg2.Error as e:
+            # A query/schema error (e.g. a renamed/missing column) gets the same clean
+            # operator message as a connection failure — not a raw SQL traceback. We
+            # never return a partial `data` (the return is the last statement in `try`).
+            raise UserError(
+                "Failed reading from the NexaCloud database — the source schema may "
+                "have changed. Underlying error:\n%s" % e)
         finally:
             conn.close()
 
@@ -127,8 +147,10 @@ class FusionBillingImportWizard(models.TransientModel):
                 partner_by_user[str(u["id"])] = link.partner_id
                 self._bump(summary, created, "partners")
             except Exception as e:  # noqa: BLE001 - per-row isolation
+                _logger.exception("NexaCloud import: user row %s failed", u.get("id"))
                 summary["failed"].append(
-                    {"kind": "user", "id": str(u.get("id")), "error": str(e)})
+                    {"kind": "user", "id": str(u.get("id")),
+                     "error": "%s: %s" % (type(e).__name__, e)})
 
         for p in data.get("plans", []):
             try:
@@ -137,8 +159,10 @@ class FusionBillingImportWizard(models.TransientModel):
                 plan_ctx_by_id[str(p["id"])] = ctx
                 self._bump(summary, created, "plans")
             except Exception as e:  # noqa: BLE001
+                _logger.exception("NexaCloud import: plan row %s failed", p.get("id"))
                 summary["failed"].append(
-                    {"kind": "plan", "id": str(p.get("id")), "error": str(e)})
+                    {"kind": "plan", "id": str(p.get("id")),
+                     "error": "%s: %s" % (type(e).__name__, e)})
 
         for s in data.get("subscriptions", []):
             partner = partner_by_user.get(str(s.get("user_id") or ""))
@@ -154,8 +178,10 @@ class FusionBillingImportWizard(models.TransientModel):
                         service, partner, ctx, recurrence_plans, s)
                 self._bump(summary, created, "subscriptions")
             except Exception as e:  # noqa: BLE001
+                _logger.exception("NexaCloud import: subscription row %s failed", s.get("id"))
                 summary["failed"].append(
-                    {"kind": "subscription", "id": str(s.get("id")), "error": str(e)})
+                    {"kind": "subscription", "id": str(s.get("id")),
+                     "error": "%s: %s" % (type(e).__name__, e)})
 
         _logger.info("NexaCloud import summary: %s", summary)
         return summary
@@ -234,8 +260,12 @@ class FusionBillingImportWizard(models.TransientModel):
         Charge = self.env["fusion.billing.charge"]
         plan_code = str(prow["id"])
         name = prow.get("name") or plan_code
-        price_monthly = float(prow.get("price_monthly") or 0.0)
-        price_yearly = float(prow.get("price_yearly") or 0.0)
+        # Preserve NULL vs 0.0: a missing price must NOT silently become a $0 line.
+        # The subscription import raises on a missing price for its cycle (-> failed[]).
+        raw_monthly = prow.get("price_monthly")
+        raw_yearly = prow.get("price_yearly")
+        price_monthly = float(raw_monthly) if raw_monthly is not None else None
+        price_yearly = float(raw_yearly) if raw_yearly is not None else None
         created = False
 
         sub_code = "NC-PLAN-%s" % plan_code
@@ -244,7 +274,7 @@ class FusionBillingImportWizard(models.TransientModel):
             sub_product = Product.create({
                 "name": "NexaCloud %s" % name, "default_code": sub_code,
                 "type": "service", "recurring_invoice": True,
-                "list_price": price_monthly})
+                "list_price": price_monthly or 0.0})
             created = True
 
         ov_code = "NC-CPU-OVG-%s" % plan_code
@@ -261,7 +291,10 @@ class FusionBillingImportWizard(models.TransientModel):
             "price_per_unit": CPU_RATE_PER_CORE_HOUR,
             "unit_batch": CPU_SECONDS_PER_CORE_HOUR,
             "charge_model": "standard",
-            # plan_id intentionally omitted (NULL) — shadow safety guarantee #3
+            # Shadow safety guarantee #3: plan_id MUST stay NULL so the rating cron
+            # never auto-mutates order lines. Set it explicitly (not just omitted) so a
+            # re-run re-asserts NULL even if someone set it on the charge between runs.
+            "plan_id": False,
         }
         charge = Charge.search(
             [("plan_code", "=", plan_code), ("metric_id", "=", metric.id)], limit=1)
@@ -280,20 +313,29 @@ class FusionBillingImportWizard(models.TransientModel):
         SaleOrder = self.env["sale.order"]
         SaleOrderLine = self.env["sale.order.line"]
         sub_ext = str(srow["id"])
-        cycle = (srow.get("billing_cycle") or "monthly").lower()
+        cycle = (srow.get("billing_cycle") or "").strip().lower()
+        if cycle not in ("monthly", "yearly"):
+            raise UserError(
+                "Subscription %s has an unrecognized billing_cycle %r — cannot pick a "
+                "plan/price." % (sub_ext, srow.get("billing_cycle")))
         rec_plan = recurrence_plans["yearly"] if cycle == "yearly" else recurrence_plans["monthly"]
         price = plan_ctx["price_yearly"] if cycle == "yearly" else plan_ctx["price_monthly"]
+        if price is None:
+            raise UserError(
+                "Subscription %s is billed %s but its plan has no %s price." % (
+                    sub_ext, cycle, cycle))
         product = plan_ctx["sub_product"]
-        order_vals = {
-            "partner_id": partner.id, "plan_id": rec_plan.id,
-            "x_fc_nexacloud_subscription_id": sub_ext,
+        # x_fc_* are always (re-)written; identity fields (partner_id/plan_id/order_line)
+        # are set ONLY at creation, so a re-run never rewrites immutable fields on an
+        # order that may since have been confirmed.
+        shadow_vals = {
             "x_fc_nexacloud_deployment_id": str(srow.get("deployment_id") or ""),
             "x_fc_billing_service_id": service.id, "x_fc_shadow": True,
         }
         existing = SaleOrder.search(
             [("x_fc_nexacloud_subscription_id", "=", sub_ext)], limit=1)
         if existing:
-            existing.write(order_vals)
+            existing.write(shadow_vals)
             line = existing.order_line.filtered(lambda l: l.product_id == product)
             line_vals = {"product_uom_qty": 1, "price_unit": price}
             if line:
@@ -304,9 +346,13 @@ class FusionBillingImportWizard(models.TransientModel):
             order = existing
             created = False
         else:
-            order_vals["order_line"] = [(0, 0, {
-                "product_id": product.id, "product_uom_qty": 1, "price_unit": price})]
-            order = SaleOrder.create(order_vals)
+            order = SaleOrder.create({
+                "partner_id": partner.id, "plan_id": rec_plan.id,
+                "x_fc_nexacloud_subscription_id": sub_ext,
+                "order_line": [(0, 0, {
+                    "product_id": product.id, "product_uom_qty": 1, "price_unit": price})],
+                **shadow_vals,
+            })
             created = True
         # guarantee the explicit price stuck (a pricelist compute may have overwritten it)
         line = order.order_line.filtered(lambda l: l.product_id == product)

fusion_helpdesk_central/__manifest__.py

@@ -3,7 +3,7 @@
 # License OPL-1
 {
     'name': 'Fusion Helpdesk Central — Client API Keys',
-    'version': '19.0.2.2.0',
+    'version': '19.0.2.3.0',
     'category': 'Productivity',
     'summary': 'Admin UI on the central Odoo for issuing per-client API '
                'keys used by fusion_helpdesk client deployments.',

fusion_helpdesk_central/models/engagement_wizard.py

@@ -64,11 +64,20 @@ class FusionHelpdeskEngagementWizard(models.TransientModel):
         help='One-line note from you, prepended above the AI summary in the '
              'email body. Optional. Skip if the summary speaks for itself.',
     )
+    findings = fields.Text(
+        string='Your Findings',
+        help='Your engineering analysis: scope, limitations, recommended '
+             'approach, effort, risks — anything the original reporter '
+             'would not have known. The AI weighs these MORE HEAVILY than '
+             'the ticket description when generating the owner summary. '
+             'Optional but strongly recommended: without it, the summary '
+             'is just the AI restating the user\'s report.',
+    )
     ai_summary = fields.Text(
-        string='AI Summary',
-        help='OpenAI-generated brief. Edit before sending if you want to '
-             'tweak the framing. Empty? The wizard fell back to manual — '
-             'type your own brief, send normally.',
+        string='Summary to Send',
+        help='Brief shown to the owner in the approval email. Either '
+             'generated from your findings + ticket (click "Generate '
+             'Summary") or written by hand. Edit freely before sending.',
     )
 
     owner_email_display = fields.Char(
@@ -129,34 +138,35 @@ class FusionHelpdeskEngagementWizard(models.TransientModel):
         return vals
 
     def _default_get_single(self, vals, ticket_id):
+        # No AI on open — user writes findings first, then clicks
+        # "Generate Summary" to fire OpenAI. Summary starts empty so the
+        # view's Send button can be disabled until either Generate runs
+        # or the user types one manually.
         ticket = self.env['helpdesk.ticket'].browse(ticket_id)
         if not ticket.exists():
             raise UserError(_('Ticket %s no longer exists.') % ticket_id)
         self._validate_engagement_target(ticket)
-        summary = self._generate_summary(ticket)
         vals.update({
             'ticket_id': ticket.id,
-            'ai_summary': summary,
-            'ai_unavailable': not bool(summary),
+            'ai_summary': '',
+            'findings': '',
+            'ai_unavailable': False,
         })
         return vals
 
     def _default_get_bulk(self, vals, ticket_ids):
+        # Same as single: no AI on open. User fills findings per ticket
+        # then hits "Generate Summary" on each line (or "Generate All" —
+        # see action_generate_all_summaries).
         tickets = self.env['helpdesk.ticket'].browse(ticket_ids).exists()
         self._validate_bulk_targets(tickets)
-        # One summary per ticket, fanned out in parallel so the modal doesn't
-        # block for N * 15s. If the fan-out itself times out we still open
-        # the wizard — the user just has to fill in summaries manually.
-        summaries = self._generate_summaries_parallel(tickets)
-        any_ok = any(s for s in summaries.values())
         vals.update({
             'ticket_ids': [(6, 0, tickets.ids)],
             'line_ids': [
-                (0, 0, {'ticket_id': t.id,
-                        'ai_summary': summaries.get(t.id, '')})
+                (0, 0, {'ticket_id': t.id, 'findings': '', 'ai_summary': ''})
                 for t in tickets
             ],
-            'ai_unavailable': not any_ok,
+            'ai_unavailable': False,
         })
         return vals
 
@@ -227,9 +237,13 @@ class FusionHelpdeskEngagementWizard(models.TransientModel):
             msg_data,
         )
 
-    def _generate_summary(self, ticket):
+    def _generate_summary(self, ticket, findings=''):
         """Single-ticket summary. Returns '' on any failure — the wizard
-        treats empty as "AI unavailable" and shows the manual fallback."""
+        treats empty as "AI unavailable" and shows the manual fallback.
+
+        `findings` is the user's free-text analysis from the wizard. The
+        prompt explicitly tells the model to weight it more than the
+        original user report — see SUMMARY_PROMPT in utils.py."""
         ICP = self.env['ir.config_parameter'].sudo()
         api_key = (ICP.get_param(
             'fusion_helpdesk_central.openai_api_key') or '').strip()
@@ -238,35 +252,39 @@ class FusionHelpdeskEngagementWizard(models.TransientModel):
         model = (ICP.get_param(
             'fusion_helpdesk_central.openai_model') or 'gpt-4o-mini').strip()
         name, desc, msgs = self._summary_inputs(ticket)
-        prompt = truncate_for_openai(build_summary_prompt(name, desc, msgs))
+        prompt = truncate_for_openai(
+            build_summary_prompt(name, desc, msgs, findings=findings)
+        )
         return call_openai_chat(api_key, model, prompt)
 
-    def _generate_summaries_parallel(self, tickets):
+    def _generate_summaries_parallel(self, ticket_findings):
         """{ticket_id: summary_or_empty} for the bulk wizard.
 
-        Submits N calls in parallel via a thread pool. Each call has its own
-        15s timeout; the whole batch is capped at _BULK_AI_TIMEOUT so a slow
-        single call doesn't hold up the rest. Anything still pending at the
-        cap returns ''."""
+        `ticket_findings` is a dict {ticket_id: (ticket_recordset, findings_str)}
+        so each parallel call uses its own per-ticket findings.
+        """
         ICP = self.env['ir.config_parameter'].sudo()
         api_key = (ICP.get_param(
             'fusion_helpdesk_central.openai_api_key') or '').strip()
         if not api_key:
-            return {t.id: '' for t in tickets}
+            return {tid: '' for tid in ticket_findings}
         model = (ICP.get_param(
             'fusion_helpdesk_central.openai_model') or 'gpt-4o-mini').strip()
 
         # Build inputs serially (DB-bound, fast) before fanning out the
         # HTTP calls in parallel.
         inputs = {}
-        for t in tickets:
-            name, desc, msgs = self._summary_inputs(t)
-            inputs[t.id] = truncate_for_openai(
-                build_summary_prompt(name, desc, msgs))
+        for tid, (ticket, findings) in ticket_findings.items():
+            name, desc, msgs = self._summary_inputs(ticket)
+            inputs[tid] = truncate_for_openai(
+                build_summary_prompt(name, desc, msgs, findings=findings))
 
-        results = {t.id: '' for t in tickets}
-        with concurrent.futures.ThreadPoolExecutor(
-                max_workers=_BULK_AI_WORKERS) as pool:
+        results = {tid: '' for tid in ticket_findings}
+        # Cancel pending futures on overall timeout so a slow OpenAI day
+        # doesn't block the wizard for ceil(N/workers) * 15s.
+        pool = concurrent.futures.ThreadPoolExecutor(
+            max_workers=_BULK_AI_WORKERS)
+        try:
             futures = {
                 pool.submit(call_openai_chat, api_key, model, p): tid
                 for tid, p in inputs.items()
@@ -288,8 +306,67 @@ class FusionHelpdeskEngagementWizard(models.TransientModel):
                     'out after %ss; remaining tickets will get empty '
                     'summaries.', _BULK_AI_TIMEOUT,
                 )
+        finally:
+            # py3.9+: cancel_futures stops queued tasks from starting;
+            # in-flight urlopen calls still finish their 15s per-call cap
+            # and drop their result on the floor.
+            pool.shutdown(wait=False, cancel_futures=True)
         return results
 
+    # ------------------------------------------------------------------
+    # User-driven AI trigger — explicit "Generate" buttons in the view.
+    # ------------------------------------------------------------------
+    def action_generate_summary(self):
+        """Single mode: fire OpenAI with the current findings, drop the
+        result into ai_summary. Returns an action to keep the wizard open
+        (replaces the current view instead of closing it).
+        """
+        self.ensure_one()
+        if self.mode != 'single' or not self.ticket_id:
+            raise UserError(_(
+                'Generate Summary only works in single-ticket mode. Use '
+                'the per-line Generate buttons on the bulk wizard.'
+            ))
+        summary = self._generate_summary(
+            self.ticket_id, findings=self.findings or '',
+        )
+        self.ai_summary = summary
+        self.ai_unavailable = not bool(summary)
+        return {
+            'type': 'ir.actions.act_window',
+            'res_model': 'fusion.helpdesk.engagement.wizard',
+            'res_id': self.id,
+            'view_mode': 'form',
+            'target': 'new',
+        }
+
+    def action_generate_all_summaries(self):
+        """Bulk mode: fire OpenAI per-ticket in parallel using each line's
+        own findings. Lines that already have a non-empty ai_summary get
+        regenerated too (the user clicked the button — they meant it).
+        """
+        self.ensure_one()
+        if self.mode != 'bulk' or not self.line_ids:
+            raise UserError(_(
+                'Generate All only works in bulk mode.'
+            ))
+        ticket_findings = {
+            line.ticket_id.id: (line.ticket_id, line.findings or '')
+            for line in self.line_ids
+        }
+        results = self._generate_summaries_parallel(ticket_findings)
+        for line in self.line_ids:
+            line.ai_summary = results.get(line.ticket_id.id, '')
+        any_ok = any(results.values())
+        self.ai_unavailable = not any_ok
+        return {
+            'type': 'ir.actions.act_window',
+            'res_model': 'fusion.helpdesk.engagement.wizard',
+            'res_id': self.id,
+            'view_mode': 'form',
+            'target': 'new',
+        }
+
     # ------------------------------------------------------------------
     # Send: write engagement state + queue mail
     # ------------------------------------------------------------------
@@ -374,7 +451,14 @@ class FusionHelpdeskEngagementWizardLine(models.TransientModel):
     ticket_name = fields.Char(
         related='ticket_id.name', readonly=True,
     )
-    ai_summary = fields.Text(
-        string='AI Summary',
-        help='Per-ticket summary — edit before send.',
+    findings = fields.Text(
+        string='Your Findings',
+        help='Per-ticket engineering findings. The Generate button on this '
+             'line uses these as the most authoritative input for the AI '
+             'summary.',
+    )
+    ai_summary = fields.Text(
+        string='Summary to Send',
+        help='Per-ticket summary — generated from findings + ticket, or '
+             'written by hand. Edit before send.',
     )

fusion_helpdesk_central/tests/test_engagement.py

@@ -172,40 +172,98 @@ class TestEngagementReset(TestEngagementBase):
 @tagged('post_install', '-at_install', 'fusion_helpdesk_central')
 class TestEngagementWizard(TestEngagementBase):
 
-    def test_single_send_via_wizard(self):
+    def test_default_get_does_not_fire_openai(self):
+        # Two-step flow: opening the wizard must NOT call OpenAI — the user
+        # has to write findings first, then click Generate. Wrap call_openai_chat
+        # in a mock that fails the test if invoked.
+        from unittest.mock import patch
         t = self._make_ticket()
-        with _patch_openai():
+        called = {'n': 0}
+        def _spy(*a, **kw):
+            called['n'] += 1
+            return 'should-not-appear'
+        with patch(
+            'odoo.addons.fusion_helpdesk_central.models.engagement_wizard.'
+            'call_openai_chat',
+            side_effect=_spy,
+        ):
             wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
                 default_ticket_id=t.id,
-                active_id=t.id,
-                active_model='helpdesk.ticket',
             ).create({})
+        self.assertEqual(called['n'], 0,
+                         'default_get must not auto-fire OpenAI.')
+        self.assertEqual(wizard.ai_summary, '')
+        self.assertEqual(wizard.findings, '')
+
+    def test_single_send_via_wizard(self):
+        t = self._make_ticket()
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_id=t.id,
+            active_id=t.id,
+            active_model='helpdesk.ticket',
+        ).create({})
         self.assertEqual(wizard.mode, 'single')
+        # User writes findings, clicks Generate Summary
+        wizard.findings = 'Scope is small; ~2 hours of work.'
+        with _patch_openai():
+            wizard.action_generate_summary()
         self.assertIn('summary bullet', wizard.ai_summary)
         wizard.personal_note = 'please review'
         result = wizard.action_send()
-        # action returns the standard close-modal action
         self.assertEqual(result.get('type'), 'ir.actions.act_window_close')
         self.assertEqual(t.x_fc_engagement_state, 'pending')
         self.assertEqual(t.x_fc_engagement_email, 'owner@testclient.com')
         self.assertTrue(t.x_fc_engagement_token)
 
-    def test_single_send_uses_current_client_key_owner(self):
-        # The wizard must read the FRESH owner contact from client_key,
-        # not a stale snapshot — if the client_key is updated between
-        # default_get and Send, Send wins.
+    def test_send_with_manual_summary_no_ai(self):
+        # Skipping Generate altogether: user types the summary by hand.
+        # Send should work without ever invoking OpenAI.
         t = self._make_ticket()
-        with _patch_openai():
-            wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
-                default_ticket_id=t.id,
-            ).create({})
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_id=t.id,
+        ).create({})
+        wizard.ai_summary = 'Manually written brief; no AI involved.'
+        wizard.action_send()
+        self.assertEqual(t.x_fc_engagement_state, 'pending')
+        self.assertEqual(t.x_fc_ai_summary,
+                         'Manually written brief; no AI involved.')
+
+    def test_generate_summary_passes_findings_to_ai(self):
+        # Pin the prompt-construction wiring: findings field must reach
+        # build_summary_prompt, which must reach call_openai_chat.
+        from unittest.mock import patch
+        t = self._make_ticket()
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_id=t.id,
+        ).create({})
+        wizard.findings = 'UNIQUE-FINDINGS-MARKER-XYZ'
+        captured = {}
+        def _spy(api_key, model, prompt, timeout=15):
+            captured['prompt'] = prompt
+            return '• generated'
+        with patch(
+            'odoo.addons.fusion_helpdesk_central.models.engagement_wizard.'
+            'call_openai_chat',
+            side_effect=_spy,
+        ):
+            wizard.action_generate_summary()
+        self.assertIn('UNIQUE-FINDINGS-MARKER-XYZ', captured['prompt'])
+
+    def test_single_send_uses_current_client_key_owner(self):
+        # Send must read the FRESH owner contact from client_key, not a
+        # stale snapshot from default_get.
+        t = self._make_ticket()
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_id=t.id,
+        ).create({})
+        wizard.ai_summary = 'manual summary'
         self.client_key.owner_email = 'changed@testclient.com'
         wizard.action_send()
         self.assertEqual(t.x_fc_engagement_email, 'changed@testclient.com')
 
     def test_wizard_rejects_ticket_without_client_label(self):
         t = self._make_ticket(x_fc_client_label=False)
-        with _patch_openai(), self.assertRaises(UserError):
+        with self.assertRaises(UserError):
             self.env['fusion.helpdesk.engagement.wizard'].with_context(
                 default_ticket_id=t.id,
             ).create({})
@@ -213,17 +271,18 @@ class TestEngagementWizard(TestEngagementBase):
     def test_wizard_rejects_when_owner_contact_missing(self):
         self.client_key.write({'owner_email': False, 'owner_name': False})
         t = self._make_ticket()
-        with _patch_openai(), self.assertRaises(UserError):
+        with self.assertRaises(UserError):
             self.env['fusion.helpdesk.engagement.wizard'].with_context(
                 default_ticket_id=t.id,
             ).create({})
 
-    def test_wizard_marks_ai_unavailable_when_summary_empty(self):
+    def test_generate_marks_ai_unavailable_when_empty(self):
         t = self._make_ticket()
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_id=t.id,
+        ).create({})
         with _patch_openai(return_value=''):
-            wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
-                default_ticket_id=t.id,
-            ).create({})
+            wizard.action_generate_summary()
         self.assertTrue(wizard.ai_unavailable)
         self.assertEqual(wizard.ai_summary, '')
 
@@ -231,34 +290,37 @@ class TestEngagementWizard(TestEngagementBase):
         ts = self.env['helpdesk.ticket']
         for i in range(3):
             ts |= self._make_ticket(name='[TESTCLIENT] Bug %s' % i)
-        with _patch_openai():
-            wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
-                default_ticket_ids=ts.ids,
-                active_ids=ts.ids,
-                active_model='helpdesk.ticket',
-                fhc_bulk=True,
-            ).create({})
+        wizard = self.env['fusion.helpdesk.engagement.wizard'].with_context(
+            default_ticket_ids=ts.ids,
+            active_ids=ts.ids,
+            active_model='helpdesk.ticket',
+            fhc_bulk=True,
+        ).create({})
         self.assertEqual(wizard.mode, 'bulk')
         self.assertEqual(len(wizard.line_ids), 3)
+        # User fills findings + generates summaries (or writes by hand)
+        for line in wizard.line_ids:
+            line.findings = 'Scope: small. Effort: low.'
+        with _patch_openai():
+            wizard.action_generate_all_summaries()
+        for line in wizard.line_ids:
+            self.assertIn('summary bullet', line.ai_summary)
         wizard.action_send()
         for t in ts:
             self.assertEqual(t.x_fc_engagement_state, 'pending')
             self.assertTrue(t.x_fc_engagement_token)
-            # Each ticket must have its OWN token
         tokens = {t.x_fc_engagement_token for t in ts}
         self.assertEqual(len(tokens), 3)
 
     def test_bulk_rejects_mixed_clients(self):
         t1 = self._make_ticket()
-        # Need another client_key for the mix to be valid otherwise the
-        # owner-contact check fires first.
         self.env['fusion.helpdesk.client.key'].create({
             'client_label': 'OTHERCLIENT',
             'owner_email': 'other@x.com', 'owner_name': 'Other',
         })
         t2 = self._make_ticket(
             name='[OTHERCLIENT] x', x_fc_client_label='OTHERCLIENT')
-        with _patch_openai(), self.assertRaises(UserError):
+        with self.assertRaises(UserError):
             self.env['fusion.helpdesk.engagement.wizard'].with_context(
                 default_ticket_ids=[t1.id, t2.id],
                 fhc_bulk=True,
@@ -268,7 +330,7 @@ class TestEngagementWizard(TestEngagementBase):
         t1 = self._make_ticket()
         t1._fc_reset_engagement('o@x.com', 'Owner', '')   # already pending
         t2 = self._make_ticket(name='[TESTCLIENT] B')
-        with _patch_openai(), self.assertRaises(UserError):
+        with self.assertRaises(UserError):
             self.env['fusion.helpdesk.engagement.wizard'].with_context(
                 default_ticket_ids=[t1.id, t2.id],
                 fhc_bulk=True,

fusion_helpdesk_central/tests/test_utils.py

@@ -54,6 +54,19 @@ class TestBuildSummaryPrompt(TransactionCase):
         # survives; '(empty)' marker keeps the model honest.
         self.assertIn('(empty)', prompt)
 
+    def test_findings_included_in_prompt(self):
+        # Findings is the support engineer's analysis — must reach OpenAI.
+        prompt = build_summary_prompt(
+            't', 'd', [], findings='Scope is small. ~2h of work.',
+        )
+        self.assertIn('Scope is small. ~2h of work.', prompt)
+
+    def test_findings_absent_uses_explicit_marker(self):
+        # Empty findings collapses to an explicit marker so the model
+        # doesn't read a blank section as "missing".
+        prompt = build_summary_prompt('t', 'd', [], findings='')
+        self.assertIn('none provided', prompt)
+
 
 @tagged('post_install', '-at_install', 'fusion_helpdesk_central')
 class TestTruncateForOpenAI(TransactionCase):

fusion_helpdesk_central/utils.py

@@ -23,22 +23,35 @@ _logger = logging.getLogger(__name__)
 SUMMARY_PROMPT = """You are summarising a customer support ticket for a busy executive
 who needs to decide whether to approve the work.
 
+The support engineer has reviewed this ticket and written their findings
+below (scope, limitations, recommended approach, anything the original
+reporter wouldn't have known). Treat those findings as authoritative —
+they reflect the actual engineering reality, not just the user's
+description of the problem. If the findings contradict the original
+report, side with the findings and call out the gap.
+
 Output rules:
 - 4-6 short bullet points, plain text (no markdown).
-- First bullet: the ask, in one sentence.
-- Second bullet: the business impact if approved.
-- Third bullet: the business impact if NOT approved (or "none material").
-- Optional bullets: cost / effort signals if any are mentioned.
+- First bullet: the ask, in one sentence, framed in the support
+  engineer's terms (not just a paraphrase of the original report).
+- Second bullet: the support engineer's recommendation, if they made one.
+- Third bullet: business impact if approved.
+- Fourth bullet: business impact if NOT approved (or "none material").
+- Optional bullets: scope / effort / risk signals from the findings.
 - Final bullet: open questions the approver should think about.
-- Do not invent facts. If the thread doesn't say, write "not stated".
+- Do not invent facts. If the findings or thread don't say, write
+  "not stated".
 - No greetings, no sign-offs, no preamble.
 
 Ticket title: {name}
-Original report:
+Original report from the user:
 {description_plain}
 
-Replies so far:
+Replies in the ticket thread:
 {messages_plain}
+
+Support engineer's findings (your most important input):
+{findings_plain}
 """
 
 # Bound the prompt sent to OpenAI so a runaway thread doesn't blow cost or
@@ -47,13 +60,17 @@ Replies so far:
 OPENAI_PROMPT_MAX_CHARS = 8000
 
 
-def build_summary_prompt(ticket_name, description_plain, messages):
-    """Render SUMMARY_PROMPT with ticket + chatter content.
+def build_summary_prompt(ticket_name, description_plain, messages,
+                         findings=''):
+    """Render SUMMARY_PROMPT with ticket + chatter + support findings.
 
     `messages` is a list of dicts with at minimum {author, date, body_plain}.
-    We render one line per message so the model can attribute who said what.
-    Empty inputs collapse to "(none)" so the prompt never has bare blanks
-    that the model could interpret as a missing section.
+    `findings` is the support engineer's free-text notes from the wizard —
+    their scope/effort/recommendation that the AI should weight more
+    heavily than the original user description.
+
+    Empty inputs collapse to explicit markers so the prompt never has
+    bare blanks the model could misread as missing sections.
     """
     name = (ticket_name or '').strip() or '(untitled)'
     desc = (description_plain or '').strip() or '(no description)'
@@ -67,8 +84,13 @@ def build_summary_prompt(ticket_name, description_plain, messages):
         msgs = '\n---\n'.join(lines)
     else:
         msgs = '(no replies yet)'
+    findings_text = (findings or '').strip() or (
+        '(none provided — base the summary on the user\'s report and '
+        'thread alone)'
+    )
     return SUMMARY_PROMPT.format(
         name=name, description_plain=desc, messages_plain=msgs,
+        findings_plain=findings_text,
     )
 
 

fusion_helpdesk_central/views/engagement_wizard_views.xml

@@ -29,28 +29,65 @@
                      invisible="not ai_unavailable">
                     <strong>AI summary unavailable.</strong>
                     OpenAI didn't return a summary (no API key set, rate limit,
-                    or network error). Write a quick brief below before sending —
-                    everything else still works.
+                    or network error). Write a quick brief in the Summary
+                    field below before sending — everything else still works.
                 </div>
 
-                <!-- Single mode: one summary field for the one ticket. -->
+                <!--
+                    Single mode. The user fills findings first, clicks
+                    Generate Summary, reviews the AI output, edits if needed,
+                    then Sends. We deliberately don't auto-fire OpenAI on
+                    open — the AI needs the engineer's analysis to be useful.
+                -->
                 <group invisible="mode != 'single'">
                     <field name="personal_note"
                            placeholder="One-line note that appears above the summary in the email…"/>
-                    <field name="ai_summary" string="Summary to send"
-                           placeholder="Bullet-point summary that the owner will read first." />
+                </group>
+                <group invisible="mode != 'single'" string="1. Your Findings">
+                    <field name="findings" nolabel="1"
+                           placeholder="What did your investigation surface? Scope, limitations, recommended approach, effort, risks — anything the owner needs to know that the original reporter wouldn't have."/>
+                </group>
+                <group invisible="mode != 'single'" string="2. Summary to Send">
+                    <div class="o_row" colspan="2">
+                        <button name="action_generate_summary" type="object"
+                                string="Generate Summary from Findings"
+                                icon="fa-magic"
+                                class="btn-secondary"/>
+                        <span class="o_form_label text-muted ms-2">
+                            ← click after writing your findings, then review &amp; edit below
+                        </span>
+                    </div>
+                    <field name="ai_summary" nolabel="1"
+                           placeholder="Click Generate Summary above, or write the brief yourself. The owner reads this first."/>
                 </group>
 
-                <!-- Bulk mode: per-ticket lines, each with its own summary. -->
+                <!--
+                    Bulk mode: same two-step flow per ticket. Findings +
+                    Summary editable inline; a single Generate All button
+                    fans out OpenAI in parallel using each line's findings.
+                -->
                 <group invisible="mode != 'bulk'">
                     <field name="personal_note"
                            placeholder="One-line note that appears once at the top of the combined email…"/>
                 </group>
+                <div class="o_row" colspan="2"
+                     invisible="mode != 'bulk'">
+                    <button name="action_generate_all_summaries" type="object"
+                            string="Generate All Summaries"
+                            icon="fa-magic"
+                            class="btn-secondary"/>
+                    <span class="o_form_label text-muted ms-2">
+                        ← fill in findings per row first, then click to regenerate all summaries in parallel
+                    </span>
+                </div>
                 <field name="line_ids" invisible="mode != 'bulk'" nolabel="1">
                     <list editable="bottom" create="0" delete="0">
                         <field name="ticket_id" readonly="1"/>
                         <field name="ticket_name" readonly="1"/>
-                        <field name="ai_summary"/>
+                        <field name="findings"
+                               placeholder="Your engineering findings for this ticket…"/>
+                        <field name="ai_summary"
+                               placeholder="(empty — click Generate All above, or write by hand)"/>
                     </list>
                 </field>
 

fusion_repairs/__manifest__.py

@@ -4,7 +4,7 @@
 
 {
     'name': 'Fusion Repairs',
-    'version': '19.0.2.2.1',
+    'version': '19.0.2.2.3',
     'category': 'Inventory/Repairs',
     'summary': 'Guided medical equipment repair intake, dispatch, maintenance, and self-service portal',
     'description': """

fusion_repairs/static/src/components/flowchart_designer/flowchart_designer.js

@@ -78,6 +78,21 @@ export class FlowchartDesigner extends Component {
         // Per-selected-node Wysiwyg editor instance captured via onLoad.
         // Keyed by dfId so switching nodes correctly reads the right editor.
         this.wysiwygEditors = {};
+
+        // Drawflow init + chart load happen on mount; cleanup on unmount.
+        // These onMounted / onWillUnmount hooks MUST stay inside setup() -
+        // they are not class methods. Leaving them outside throws
+        // "Unexpected token (" because JS parses bare function calls in a
+        // class body as malformed method declarations.
+        onMounted(async () => {
+            await Promise.all([loadJS(DRAWFLOW_JS), loadCSS(DRAWFLOW_CSS)]);
+            this._initDrawflow();
+            await this._loadChart();
+        });
+
+        onWillUnmount(() => {
+            try { this.editor?.clear(); } catch {}
+        });
     }
 
     // ------------------------------------------------------------------
@@ -124,17 +139,6 @@ export class FlowchartDesigner extends Component {
         this.state.sourceMode = !this.state.sourceMode;
     }
 
-        onMounted(async () => {
-            await Promise.all([loadJS(DRAWFLOW_JS), loadCSS(DRAWFLOW_CSS)]);
-            this._initDrawflow();
-            await this._loadChart();
-        });
-
-        onWillUnmount(() => {
-            try { this.editor?.clear(); } catch {}
-        });
-    }
-
     _initDrawflow() {
         // eslint-disable-next-line no-undef
         this.editor = new Drawflow(this.canvasRef.el);

fusion_repairs/static/src/components/flowchart_designer/flowchart_designer.xml

@@ -91,7 +91,7 @@
                                 <div class="fr-wysiwyg-shell">
                                     <Wysiwyg t-key="'wysiwyg-' + state.selectedNodeId"
                                              config="wysiwygConfig"
-                                             onLoad="onWysiwygLoad.bind(this)"
+                                             onLoad.bind="onWysiwygLoad"
                                              contentClass="'fr-wysiwyg-content'"/>
                                 </div>
                                 <div class="form-text">Bold, italic, lists, links and inline images supported. Click <em>HTML Source</em> to paste raw markup.</div>