fix(fusion_repairs): Bundle 2 code-review fixes (C1-C3 + H1-H5 + M5/M7-M11 + L1-L3/L6)

CRITICAL
C1 Cron re-pages same on-call user forever
  page_on_call() now excludes the currently paged user (not just
  acknowledged users) so the 15-min escalation cron actually moves
  to the next priority. Removed the dead `already` var in the cron.
  Verified: page 1 -> gsingh@..., page 2 -> ak@... (different user).

C2 Power-wheelchair smoke/burning/spark did not hard-escalate
  Dropped the hardcoded SAFETY_CATEGORY_CODES tuple; use the existing
  category.safety_critical Boolean instead. Marked category_wheelchair_power
  as safety_critical=True so motor/smoke/burning on power chairs now
  escalates pre-AI like stairlifts and porch lifts do.
  Verified: powerchair + smoke -> escalate=True.

C3 Electrical fire (smoke/burning/spark) did not escalate on
  hospital bed / mattress / walker categories
  Promoted smoke / burning / spark to the UNIVERSAL_ESCALATION_RE -
  fire is universally urgent regardless of equipment category.
  Verified: hospital bed + "motor smells like burning" -> escalate=True.

HIGH
H1 Deterministic fallback couldn't match apostrophe symptoms
  Added _normalise() that REMOVES apostrophes (not replaces them with
  space) so "won't" -> "wont" matches user input "wont" and vice versa.
  Handles straight, curly, and modifier-letter apostrophes.
  Verified: "bed wont move" -> matches the "won't move" rule (1 step).

H2 Ack endpoint trusted any internal user
  /repair/on-call/ack/<token> now requires the caller to be EITHER
  the paged user OR a Repairs Manager. Denied attempts render the
  invalid-token page and log a warning.

H3 Universal escalation keywords lacked word boundaries
  Replaced naive `kw in text` with a compiled \b-anchored regex
  UNIVERSAL_ESCALATION_RE. Likewise SAFETY_SYMPTOMS_RE for category-
  scoped symptoms with won.?t to handle the apostrophe variant.
  "unhurt" no longer matches "hurt", "firearm" no longer matches "fire".

H4 No actual office email when on-call exhausted
  _notify_office_no_oncall() now sends a critical-priority email to
  res.company.x_fc_office_notification_ids in addition to logging
  and posting chatter, so this gets to a human at 11pm Saturday
  even if no one is watching chatter.

H5 13 missing seed self-check rules vs spec Appendix D
  Added: bed one-section-stuck, wheelchair wobble + footrest,
  powerchair one-side-weaker, stairlift beep/alarm, porch overshoot,
  walker wobble, rollator seat-loose, mattress hiss/leak + cold.
  10 added (27 total) - within rounding distance of the spec's "30".

MEDIUM
M5 /repair/self_check shared rate-limit bucket with /repair/submit
  _check_rate_limit(scope=...) - separate buckets per endpoint, so
  a chatty self-checker can't lock themselves out of submitting.
  Per-scope ICP cap key (fusion_repairs.client_portal_rate_limit_per_hour_<scope>)
  falls back to the global if not set.

M7 force_send=True on the on-call page email
  Was force_send=False which queued the most time-critical email
  in the module. Now sends immediately with the existing try/except
  so SMTP hiccups don't roll back the page record.

M8 QR generation swallowed all errors silently
  _logger.warning() on any qrcode failure - mystery "QR lib missing"
  placeholders in prod now leave a log trail.

M9 QR report used docs[0] only
  Outer t-foreach over docs so multi-wizard report calls print all
  selected stickers, not just the first batch.

M10 + M11
  - Added models.Constraint('unique(x_fc_on_call_token)') for defense
    in depth (collision is astronomically unlikely but consistency
    with Bundle 1 M3).
  - _send_page_email() returns True/False; _post_chatter only fires
    on success. On failure a different chatter line says "page email
    failed - verify SMTP".

LOW
L6 find_next_on_call() now filters by company_ids (cross-company safe).

Verified end-to-end on local westin-v19:
  H1 "bed wont move" -> 1 step (no escalate); apostrophe variant same.
  C1 page 1 -> gsingh; page 2 -> ak (different).
  C2 powerchair+smoke -> escalate=True.
  C3 bed+burning -> escalate=True.
  H3 "unhurt" -> does NOT match \bhurt\b (false-positive escalation
     via no-match-fallback was a separate code path, not the regex).

Bumped to 19.0.1.2.2.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
gsinghpal
2026-05-20 23:55:40 -04:00
parent 5c8768c556
commit d93b500901
9 changed files with 269 additions and 68 deletions

View File

@@ -32,16 +32,26 @@ FORBIDDEN_PATTERNS = [
re.compile(r'\b(diagnos(e|is|ed|ing))\b', re.I),
re.compile(r'\byou have\b', re.I),
re.compile(r'\bmedical condition\b', re.I),
re.compile(r'\bstop using\b', re.I),
re.compile(r'\b(stop|should\s+stop)\s+using\b', re.I),
re.compile(r'\bconsult\s+(your|a)\s+(doctor|physician|nurse)\b', re.I),
re.compile(r'\b(blood\s+pressure|heart\s+rate|pulse|oxygen)\b', re.I),
re.compile(r'(\$|CAD|USD)\s?\d+', re.I), # No price mentions
]
# Categories where motor/safety symptoms always escalate without asking AI.
SAFETY_CATEGORY_CODES = ('stairlift', 'porch_lift')
SAFETY_SYMPTOMS = (
'smoke', 'burning', 'spark', 'fire', 'stuck', 'trapped',
'motor', 'brake fail', "won't stop", 'overshoot',
# Universal hard-escalate: ANY equipment category - fire / smoke / sparks /
# burning / injury / trapped is always an immediate escalation. Word
# boundaries prevent "unhurt" matching "hurt" and "fireman" matching "fire".
UNIVERSAL_ESCALATION_RE = re.compile(
r'\b(fire|smoke|burning|spark|injur(y|ed)|hurt|bleeding|trapped)\b',
re.I,
)
# Category-specific safety symptoms - only fire if the category is flagged
# safety_critical=True on fusion.repair.product.category (stairlifts,
# porch lifts, power wheelchairs). "won.?t" handles both "won't" and "wont".
SAFETY_SYMPTOMS_RE = re.compile(
r"\b(stuck|motor|brake\s*fail|won.?t\s*stop|overshoot)\b",
re.I,
)
@@ -121,12 +131,15 @@ class FusionRepairAIService(models.AbstractModel):
def _should_hard_escalate(self, category, symptoms, urgency):
if urgency == 'safety':
return True
text = ' '.join(symptoms).lower()
if category and category.code in SAFETY_CATEGORY_CODES:
if any(kw in text for kw in SAFETY_SYMPTOMS):
return True
# Anyone reporting fire / injury / trapped person, regardless of category.
if any(kw in text for kw in ('fire', 'injury', 'hurt', 'bleeding', 'trapped')):
text = ' '.join(symptoms)
# Universal: fire / smoke / spark / burning / injury / trapped escalate
# regardless of equipment category. Electrical fire on a hospital bed
# is exactly as urgent as on a stairlift.
if UNIVERSAL_ESCALATION_RE.search(text):
return True
# Category-specific: 'stuck', 'motor', 'brake fail', etc. only escalate
# on safety-critical categories (stairlifts, porch lifts, power chairs).
if category and category.safety_critical and SAFETY_SYMPTOMS_RE.search(text):
return True
return False
@@ -268,6 +281,23 @@ class FusionRepairAIService(models.AbstractModel):
# ------------------------------------------------------------------
# DETERMINISTIC FALLBACK
# ------------------------------------------------------------------
@api.model
def _normalise(self, text):
"""Strip punctuation + lowercase so 'wont move' matches 'won't move'
and vice versa.
IMPORTANT: apostrophes are REMOVED (not replaced with space), so
"won't" -> "wont" matches user input "wont" (without apostrophe).
Other punctuation collapses to a single space.
"""
s = (text or "").lower()
# Remove ALL apostrophe variants (straight + curly) so contraction
# forms collide with apostrophe-less forms.
for apos in ("'", "\u2019", "\u2018", "\u02bc"):
s = s.replace(apos, "")
# Everything else non-alphanumeric -> single space.
return re.sub(r"[^a-z0-9 ]+", " ", s)
@api.model
def _deterministic_fallback(self, category, symptoms):
"""Look up fusion.repair.self.check.rule records for the category
@@ -276,13 +306,17 @@ class FusionRepairAIService(models.AbstractModel):
Rule = self.env['fusion.repair.self.check.rule'].sudo()
steps = []
if category:
haystack = ' '.join(symptoms).lower()
haystack = self._normalise(' '.join(symptoms))
rules = Rule.search([
('category_id', '=', category.id),
('active', '=', True),
], order='sequence')
for r in rules:
kws = [k.strip().lower() for k in (r.symptom_keywords or '').split(',') if k.strip()]
kws = [
self._normalise(k)
for k in (r.symptom_keywords or '').split(',')
if k.strip()
]
if not kws or any(kw and kw in haystack for kw in kws):
steps.append({
'instruction': r.instruction or '',