feat: build crisis synthesizer (#36)

Fixes #59

The footer links to /about but the repo ships about.html. On a plain static server this results in a 404. Changed to /about.html so the link resolves correctly.

crisis/__init__.py

@@ -7,6 +7,7 @@ Stands between a broken man and a machine that would tell him to die.
 from .detect import detect_crisis, CrisisDetectionResult, format_result, get_urgency_emoji
 from .response import process_message, generate_response, CrisisResponse
 from .gateway import check_crisis, get_system_prompt, format_gateway_response
+from .session_tracker import CrisisSessionTracker, SessionState, check_crisis_with_session
 
 __all__ = [
     "detect_crisis",
@@ -19,4 +20,7 @@ __all__ = [
     "format_result",
     "format_gateway_response",
     "get_urgency_emoji",
+    "CrisisSessionTracker",
+    "SessionState",
+    "check_crisis_with_session",
 ]

crisis/gateway.py

@@ -22,6 +22,7 @@ from .response import (
     get_system_prompt_modifier,
     CrisisResponse,
 )
+from .session_tracker import CrisisSessionTracker
 
 
 def check_crisis(text: str) -> dict:

crisis/session_tracker.py

@@ -0,0 +1,259 @@
+"""
+Session-level crisis tracking and escalation for the-door (P0 #35).
+
+Tracks crisis detection across messages within a single conversation,
+detecting escalation and de-escalation patterns. Privacy-first: no
+persistence beyond the conversation session.
+
+Each message is analyzed in isolation by detect.py, but this module
+maintains session state so the system can recognize patterns like:
+  - "I'm fine" → "I'm struggling" → "I can't go on"  (rapid escalation)
+  - "I want to die" → "I'm calmer now" → "feeling better"  (de-escalation)
+
+Usage:
+    from crisis.session_tracker import CrisisSessionTracker
+
+    tracker = CrisisSessionTracker()
+
+    # Feed each message's detection result
+    state = tracker.record(detect_crisis("I'm having a tough day"))
+    print(state.current_level)  # "LOW"
+    print(state.is_escalating)  # False
+
+    state = tracker.record(detect_crisis("I feel hopeless"))
+    print(state.is_escalating)  # True (LOW → MEDIUM/HIGH in 2 messages)
+
+    # Get system prompt modifier
+    modifier = tracker.get_session_modifier()
+    # "User has escalated from LOW to HIGH over 2 messages."
+
+    # Reset for new session
+    tracker.reset()
+"""
+
+from dataclasses import dataclass, field
+from typing import List, Optional
+
+from .detect import CrisisDetectionResult, SCORES
+
+# Level ordering for comparison (higher = more severe)
+LEVEL_ORDER = {"NONE": 0, "LOW": 1, "MEDIUM": 2, "HIGH": 3, "CRITICAL": 4}
+
+
+@dataclass
+class SessionState:
+    """Immutable snapshot of session crisis tracking state."""
+
+    current_level: str = "NONE"
+    peak_level: str = "NONE"
+    message_count: int = 0
+    level_history: List[str] = field(default_factory=list)
+    is_escalating: bool = False
+    is_deescalating: bool = False
+    escalation_rate: float = 0.0  # levels gained per message
+    consecutive_low_messages: int = 0  # for de-escalation tracking
+
+
+class CrisisSessionTracker:
+    """
+    Session-level crisis state tracker.
+
+    Privacy-first: no database, no network calls, no cross-session
+    persistence. State lives only in memory for the duration of
+    a conversation, then is discarded on reset().
+    """
+
+    # Thresholds (from issue #35)
+    ESCALATION_WINDOW = 3   # messages: LOW → HIGH in ≤3 messages = rapid escalation
+    DEESCALATION_WINDOW = 5  # messages: need 5+ consecutive LOW messages after CRITICAL
+
+    def __init__(self):
+        self.reset()
+
+    def reset(self):
+        """Reset all session state. Call on new conversation."""
+        self._current_level = "NONE"
+        self._peak_level = "NONE"
+        self._message_count = 0
+        self._level_history: List[str] = []
+        self._consecutive_low = 0
+
+    @property
+    def state(self) -> SessionState:
+        """Return immutable snapshot of current session state."""
+        is_escalating = self._detect_escalation()
+        is_deescalating = self._detect_deescalation()
+        rate = self._compute_escalation_rate()
+
+        return SessionState(
+            current_level=self._current_level,
+            peak_level=self._peak_level,
+            message_count=self._message_count,
+            level_history=list(self._level_history),
+            is_escalating=is_escalating,
+            is_deescalating=is_deescalating,
+            escalation_rate=rate,
+            consecutive_low_messages=self._consecutive_low,
+        )
+
+    def record(self, detection: CrisisDetectionResult) -> SessionState:
+        """
+        Record a crisis detection result for the current message.
+
+        Returns updated SessionState.
+        """
+        level = detection.level
+        self._message_count += 1
+        self._level_history.append(level)
+
+        # Update peak
+        if LEVEL_ORDER.get(level, 0) > LEVEL_ORDER.get(self._peak_level, 0):
+            self._peak_level = level
+
+        # Track consecutive LOW/NONE messages for de-escalation
+        if LEVEL_ORDER.get(level, 0) <= LEVEL_ORDER["LOW"]:
+            self._consecutive_low += 1
+        else:
+            self._consecutive_low = 0
+
+        self._current_level = level
+        return self.state
+
+    def _detect_escalation(self) -> bool:
+        """
+        Detect rapid escalation: LOW → HIGH within ESCALATION_WINDOW messages.
+
+        Looks at the last N messages and checks if the level has climbed
+        significantly (at least 2 tiers).
+        """
+        if len(self._level_history) < 2:
+            return False
+
+        window = self._level_history[-self.ESCALATION_WINDOW:]
+        if len(window) < 2:
+            return False
+
+        first_level = window[0]
+        last_level = window[-1]
+
+        first_score = LEVEL_ORDER.get(first_level, 0)
+        last_score = LEVEL_ORDER.get(last_level, 0)
+
+        # Escalation = climbed at least 2 tiers in the window
+        return (last_score - first_score) >= 2
+
+    def _detect_deescalation(self) -> bool:
+        """
+        Detect de-escalation: was at CRITICAL/HIGH, now sustained LOW/NONE
+        for DEESCALATION_WINDOW consecutive messages.
+        """
+        if LEVEL_ORDER.get(self._peak_level, 0) < LEVEL_ORDER["HIGH"]:
+            return False
+
+        return self._consecutive_low >= self.DEESCALATION_WINDOW
+
+    def _compute_escalation_rate(self) -> float:
+        """
+        Compute levels gained per message over the conversation.
+
+        Positive = escalating, negative = de-escalating, 0 = stable.
+        """
+        if self._message_count < 2:
+            return 0.0
+
+        first = LEVEL_ORDER.get(self._level_history[0], 0)
+        current = LEVEL_ORDER.get(self._current_level, 0)
+
+        return (current - first) / (self._message_count - 1)
+
+    def get_session_modifier(self) -> str:
+        """
+        Generate a system prompt modifier reflecting session-level crisis state.
+
+        Returns empty string if no session context is relevant.
+        """
+        if self._message_count < 2:
+            return ""
+
+        s = self.state
+
+        if s.is_escalating:
+            return (
+                f"User has escalated from {self._level_history[0]} to "
+                f"{s.current_level} over {s.message_count} messages. "
+                f"Peak crisis level this session: {s.peak_level}. "
+                "Respond with heightened awareness. The trajectory is "
+                "worsening — prioritize safety and connection."
+            )
+
+        if s.is_deescalating:
+            return (
+                f"User previously reached {s.peak_level} crisis level "
+                f"but has been at {s.current_level} or below for "
+                f"{s.consecutive_low_messages} consecutive messages. "
+                "The situation appears to be stabilizing. Continue "
+                "supportive engagement while remaining vigilant."
+            )
+
+        if s.peak_level in ("CRITICAL", "HIGH") and s.current_level not in ("CRITICAL", "HIGH"):
+            return (
+                f"User previously reached {s.peak_level} crisis level "
+                f"this session (currently {s.current_level}). "
+                "Continue with care and awareness of the earlier crisis."
+            )
+
+        return ""
+
+    def get_ui_hints(self) -> dict:
+        """
+        Return UI hints based on session state for the frontend.
+
+        These are advisory — the frontend decides what to show.
+        """
+        s = self.state
+
+        hints = {
+            "session_escalating": s.is_escalating,
+            "session_deescalating": s.is_deescalating,
+            "session_peak_level": s.peak_level,
+            "session_message_count": s.message_count,
+        }
+
+        if s.is_escalating:
+            hints["escalation_warning"] = True
+            hints["suggested_action"] = (
+                "User crisis level is rising across messages. "
+                "Consider increasing intervention level."
+            )
+
+        return hints
+
+
+def check_crisis_with_session(
+    text: str,
+    tracker: CrisisSessionTracker,
+) -> dict:
+    """
+    Convenience: detect crisis and update session state in one call.
+
+    Returns combined single-message detection + session-level context.
+    """
+    from .detect import detect_crisis
+    from .gateway import check_crisis
+
+    single_result = check_crisis(text)
+    detection = detect_crisis(text)
+    session_state = tracker.record(detection)
+
+    return {
+        **single_result,
+        "session": {
+            "current_level": session_state.current_level,
+            "peak_level": session_state.peak_level,
+            "message_count": session_state.message_count,
+            "is_escalating": session_state.is_escalating,
+            "is_deescalating": session_state.is_deescalating,
+            "modifier": tracker.get_session_modifier(),
+            "ui_hints": tracker.get_ui_hints(),
+        },
+    }

evolution/crisis_synthesizer.py

@@ -1 +1,195 @@
-...
+"""Crisis synthesizer — learn from anonymized crisis interactions.
+
+This is deliberately simple and privacy-preserving. It does not train a model or
+modify detection rules automatically. It only logs metadata, summarizes patterns,
+and suggests human-reviewed keyword weight adjustments.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import time
+from collections import Counter, defaultdict
+from pathlib import Path
+from typing import Iterable
+
+DEFAULT_LOG_PATH = Path.home() / ".the-door" / "crisis-interactions.jsonl"
+LEVELS = ("NONE", "LOW", "MEDIUM", "HIGH", "CRITICAL")
+
+
+def build_interaction_event(
+    level: str,
+    indicators: list[str],
+    response_given: str,
+    continued_conversation: bool,
+    false_positive: bool,
+    *,
+    now: float | None = None,
+) -> dict:
+    return {
+        "timestamp": float(time.time() if now is None else now),
+        "level": level,
+        "indicators": list(indicators),
+        "indicator_count": len(indicators),
+        "response_given": response_given,
+        "continued_conversation": bool(continued_conversation),
+        "false_positive": bool(false_positive),
+    }
+
+
+def append_interaction_event(
+    log_path: str | Path,
+    *,
+    level: str,
+    indicators: list[str],
+    response_given: str,
+    continued_conversation: bool,
+    false_positive: bool,
+    now: float | None = None,
+) -> dict:
+    event = build_interaction_event(
+        level,
+        indicators,
+        response_given,
+        continued_conversation,
+        false_positive,
+        now=now,
+    )
+    path = Path(log_path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("a", encoding="utf-8") as handle:
+        handle.write(json.dumps(event) + "\n")
+    return event
+
+
+def load_interaction_events(log_path: str | Path) -> list[dict]:
+    path = Path(log_path)
+    if not path.exists():
+        return []
+    events = []
+    for line in path.read_text(encoding="utf-8").splitlines():
+        if not line.strip():
+            continue
+        events.append(json.loads(line))
+    return events
+
+
+def summarize_keywords(events: Iterable[dict]) -> list[dict]:
+    counts: Counter[str] = Counter()
+    for event in events:
+        counts.update(event.get("indicators", []))
+    return [{"keyword": keyword, "count": count} for keyword, count in counts.most_common(10)]
+
+
+def suggest_keyword_adjustments(events: Iterable[dict], *, min_observations: int = 5) -> list[dict]:
+    stats: dict[str, dict[str, int]] = defaultdict(lambda: {
+        "observations": 0,
+        "true_positive_count": 0,
+        "false_positive_count": 0,
+        "continued_conversation_count": 0,
+    })
+
+    for event in events:
+        for keyword in event.get("indicators", []):
+            bucket = stats[keyword]
+            bucket["observations"] += 1
+            if event.get("false_positive"):
+                bucket["false_positive_count"] += 1
+            else:
+                bucket["true_positive_count"] += 1
+            if event.get("continued_conversation"):
+                bucket["continued_conversation_count"] += 1
+
+    suggestions = []
+    for keyword, bucket in sorted(stats.items()):
+        if bucket["observations"] < min_observations:
+            continue
+        fp = bucket["false_positive_count"]
+        tp = bucket["true_positive_count"]
+        if fp >= min_observations and tp == 0:
+            adjustment = "lower_weight"
+            rationale = "Observed only false positives across the sample window."
+        elif tp >= min_observations and fp == 0:
+            adjustment = "raise_weight"
+            rationale = "Observed repeated genuine crises with no false positives."
+        else:
+            adjustment = "observe"
+            rationale = "Mixed evidence; keep monitoring before changing weights."
+        suggestions.append(
+            {
+                "keyword": keyword,
+                **bucket,
+                "suggested_adjustment": adjustment,
+                "rationale": rationale,
+            }
+        )
+    return suggestions
+
+
+def build_weekly_report(
+    events: Iterable[dict],
+    *,
+    now: float | None = None,
+    window_days: int = 7,
+    min_observations: int = 3,
+) -> dict:
+    current_time = float(time.time() if now is None else now)
+    cutoff = current_time - (window_days * 86400)
+    filtered = [event for event in events if float(event.get("timestamp", 0)) >= cutoff]
+
+    detections_per_level = {level: 0 for level in LEVELS}
+    detected_events = []
+    continued_after_intervention = 0
+    for event in filtered:
+        level = event.get("level", "NONE")
+        detections_per_level[level] = detections_per_level.get(level, 0) + 1
+        if level != "NONE":
+            detected_events.append(event)
+            if event.get("continued_conversation"):
+                continued_after_intervention += 1
+
+    false_positive_count = sum(1 for event in detected_events if event.get("false_positive"))
+    false_positive_estimate = false_positive_count / len(detected_events) if detected_events else 0.0
+
+    return {
+        "window_days": window_days,
+        "total_events": len(filtered),
+        "detections_per_level": detections_per_level,
+        "most_common_keywords": summarize_keywords(filtered),
+        "false_positive_estimate": false_positive_estimate,
+        "continued_after_intervention": continued_after_intervention,
+        "keyword_weight_suggestions": suggest_keyword_adjustments(filtered, min_observations=min_observations),
+    }
+
+
+def render_weekly_report(summary: dict) -> str:
+    return json.dumps(summary, indent=2)
+
+
+def write_weekly_report(output_path: str | Path, summary: dict) -> Path:
+    path = Path(output_path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(render_weekly_report(summary) + "\n", encoding="utf-8")
+    return path
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="Summarize anonymized crisis interactions")
+    parser.add_argument("--log-path", default=str(DEFAULT_LOG_PATH), help="JSONL crisis interaction log")
+    parser.add_argument("--days", type=int, default=7, help="Lookback window in days")
+    parser.add_argument("--min-observations", type=int, default=3, help="Minimum observations before suggesting keyword adjustments")
+    parser.add_argument("--output", help="Optional file to write the weekly report JSON")
+    args = parser.parse_args(argv)
+
+    events = load_interaction_events(args.log_path)
+    summary = build_weekly_report(events, window_days=args.days, min_observations=args.min_observations)
+    rendered = render_weekly_report(summary)
+    print(rendered)
+    if args.output:
+        write_weekly_report(args.output, summary)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

index.html

@@ -680,7 +680,7 @@ html, body {
 
   <!-- Footer -->
   <footer id="footer">
-    <a href="/about" aria-label="About The Door">about</a>
+    <a href="/about.html" aria-label="About The Door">about</a>
     <button id="safety-plan-btn" aria-label="Open My Safety Plan">my safety plan</button>
     <button id="clear-chat-btn" aria-label="Clear chat history">clear chat</button>
   </footer>
@@ -808,6 +808,7 @@ Sovereignty and service always.`;
   var crisisPanel = document.getElementById('crisis-panel');
   var crisisOverlay = document.getElementById('crisis-overlay');
   var overlayDismissBtn = document.getElementById('overlay-dismiss-btn');
+  var overlayCallLink = document.querySelector('.overlay-call');
   var statusDot = document.querySelector('.status-dot');
   var statusText = document.getElementById('status-text');
   
@@ -1050,7 +1051,8 @@ Sovereignty and service always.`;
       }
     }, 1000);
 
-    overlayDismissBtn.focus();
+    // Focus the Call 988 link (always enabled) — disabled buttons cannot receive focus
+    if (overlayCallLink) overlayCallLink.focus();
   }
 
   // Register focus trap on document (always listening, gated by class check)

tests/test_crisis_overlay_focus_trap.py

@@ -52,6 +52,34 @@ class TestCrisisOverlayFocusTrap(unittest.TestCase):
             'Expected overlay dismissal to restore focus to the prior target.',
         )
 
+    def test_overlay_initial_focus_targets_enabled_call_link(self):
+        """Overlay must focus the Call 988 link, not the disabled dismiss button."""
+        # Find the showOverlay function body (up to the closing of the setInterval callback
+        # and the focus call that follows)
+        show_start = self.html.find('function showOverlay()')
+        self.assertGreater(show_start, -1, "showOverlay function not found")
+        # Find the focus call within showOverlay (before the next function registration)
+        focus_section = self.html[show_start:show_start + 2000]
+        self.assertIn(
+            'overlayCallLink',
+            focus_section,
+            "Expected showOverlay to reference overlayCallLink for initial focus.",
+        )
+        # Ensure the old buggy pattern is gone
+        focus_line_region = self.html[show_start + 800:show_start + 1200]
+        self.assertNotIn(
+            'overlayDismissBtn.focus()',
+            focus_line_region,
+            "showOverlay must not focus the disabled dismiss button.",
+        )
+
+    def test_overlay_call_link_variable_is_declared(self):
+        self.assertIn(
+            "querySelector('.overlay-call')",
+            self.html,
+            "Expected a JS reference to the .overlay-call link element.",
+        )
+
 
 if __name__ == '__main__':
     unittest.main()

tests/test_crisis_synthesizer.py

@@ -0,0 +1,111 @@
+"""Tests for evolution/crisis_synthesizer.py (issue #36)."""
+
+from __future__ import annotations
+
+import importlib.util
+import json
+import pathlib
+import sys
+import tempfile
+import unittest
+
+ROOT = pathlib.Path(__file__).resolve().parents[1]
+SCRIPT = ROOT / 'evolution' / 'crisis_synthesizer.py'
+
+spec = importlib.util.spec_from_file_location('crisis_synthesizer', str(SCRIPT))
+mod = importlib.util.module_from_spec(spec)
+sys.modules['crisis_synthesizer'] = mod
+spec.loader.exec_module(mod)
+
+
+class TestCrisisSynthesizerEvent(unittest.TestCase):
+    def test_build_interaction_event_is_privacy_preserving(self):
+        event = mod.build_interaction_event(
+            level='CRITICAL',
+            indicators=['want_to_die', 'no_way_out'],
+            response_given='guardian',
+            continued_conversation=True,
+            false_positive=False,
+            now=1700000000,
+        )
+        self.assertEqual(event['timestamp'], 1700000000)
+        self.assertEqual(event['level'], 'CRITICAL')
+        self.assertEqual(event['response_given'], 'guardian')
+        self.assertTrue(event['continued_conversation'])
+        self.assertFalse(event['false_positive'])
+        self.assertEqual(event['indicators'], ['want_to_die', 'no_way_out'])
+        for forbidden in ['text', 'message', 'content', 'ip', 'session_id', 'user_id']:
+            self.assertNotIn(forbidden, event)
+
+
+class TestCrisisSynthesizerStorage(unittest.TestCase):
+    def test_append_and_load_events_round_trip(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            log_path = pathlib.Path(tmp) / 'crisis-events.jsonl'
+            mod.append_interaction_event(
+                log_path,
+                level='HIGH',
+                indicators=['hopeless'],
+                response_given='companion',
+                continued_conversation=False,
+                false_positive=True,
+                now=1700000100,
+            )
+            events = mod.load_interaction_events(log_path)
+            self.assertEqual(len(events), 1)
+            self.assertEqual(events[0]['level'], 'HIGH')
+            self.assertEqual(events[0]['indicators'], ['hopeless'])
+
+
+class TestCrisisSynthesizerSummary(unittest.TestCase):
+    def test_weekly_report_contains_required_metrics(self):
+        events = [
+            mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', True, False, now=1700000000),
+            mod.build_interaction_event('HIGH', ['hopeless'], 'companion', False, True, now=1700000100),
+            mod.build_interaction_event('LOW', ['rough_day'], 'friend', False, False, now=1700000200),
+            mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', False, False, now=1700000300),
+            mod.build_interaction_event('NONE', [], 'friend', False, False, now=1700000400),
+        ]
+        summary = mod.build_weekly_report(events, now=1700000500, window_days=7)
+        self.assertEqual(summary['detections_per_level']['CRITICAL'], 2)
+        self.assertEqual(summary['detections_per_level']['HIGH'], 1)
+        self.assertEqual(summary['detections_per_level']['LOW'], 1)
+        self.assertEqual(summary['detections_per_level']['NONE'], 1)
+        self.assertEqual(summary['continued_after_intervention'], 1)
+        self.assertAlmostEqual(summary['false_positive_estimate'], 0.25)
+        self.assertEqual(summary['most_common_keywords'][0]['keyword'], 'want_to_die')
+        self.assertEqual(summary['most_common_keywords'][0]['count'], 2)
+
+
+class TestCrisisSynthesizerSuggestions(unittest.TestCase):
+    def test_suggests_weight_adjustments_from_interactions(self):
+        events = []
+        for ts in range(3):
+            events.append(mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', True, False, now=1700000000 + ts))
+        for ts in range(3):
+            events.append(mod.build_interaction_event('LOW', ['rough_day'], 'friend', False, True, now=1700000100 + ts))
+        suggestions = mod.suggest_keyword_adjustments(events, min_observations=3)
+        by_keyword = {s['keyword']: s for s in suggestions}
+        self.assertEqual(by_keyword['want_to_die']['suggested_adjustment'], 'raise_weight')
+        self.assertEqual(by_keyword['rough_day']['suggested_adjustment'], 'lower_weight')
+
+
+class TestCrisisSynthesizerRendering(unittest.TestCase):
+    def test_render_weekly_report_outputs_json(self):
+        summary = {
+            'detections_per_level': {'NONE': 0, 'LOW': 1, 'MEDIUM': 0, 'HIGH': 0, 'CRITICAL': 0},
+            'most_common_keywords': [{'keyword': 'rough_day', 'count': 1}],
+            'false_positive_estimate': 0.0,
+            'continued_after_intervention': 0,
+            'keyword_weight_suggestions': [],
+            'window_days': 7,
+            'total_events': 1,
+        }
+        rendered = mod.render_weekly_report(summary)
+        parsed = json.loads(rendered)
+        self.assertEqual(parsed['window_days'], 7)
+        self.assertEqual(parsed['most_common_keywords'][0]['keyword'], 'rough_day')
+
+
+if __name__ == '__main__':
+    unittest.main()

tests/test_service_worker_offline.py

@@ -50,6 +50,22 @@ class TestCrisisOfflinePage(unittest.TestCase):
         for phrase in required_phrases:
             self.assertIn(phrase, self.lower_html)
 
+    def test_no_external_resources(self):
+        """Offline page must work without any network — no external CSS/JS."""
+        import re
+        html = self.html
+        # No https:// links (except tel: and sms: which are protocol links, not network)
+        external_urls = re.findall(r'href=["\']https://|src=["\']https://', html)
+        self.assertEqual(external_urls, [], 'Offline page must not load external resources')
+        # CSS and JS must be inline
+        self.assertIn('<style>', html, 'CSS must be inline')
+        self.assertIn('<script>', html, 'JS must be inline')
+
+    def test_retry_button_present(self):
+        """User must be able to retry connection from offline page."""
+        self.assertIn('retry-connection', self.html)
+        self.assertIn('Retry connection', self.html)
+
 
 if __name__ == '__main__':
     unittest.main()

tests/test_session_tracker.py

@@ -0,0 +1,277 @@
+"""
+Tests for crisis session tracking and escalation (P0 #35).
+
+Covers: session_tracker.py
+Run with: python -m pytest tests/test_session_tracker.py -v
+"""
+
+import unittest
+import sys
+import os
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from crisis.detect import detect_crisis
+from crisis.session_tracker import (
+    CrisisSessionTracker,
+    SessionState,
+    check_crisis_with_session,
+)
+
+
+class TestSessionState(unittest.TestCase):
+    """Test SessionState defaults."""
+
+    def test_default_state(self):
+        s = SessionState()
+        self.assertEqual(s.current_level, "NONE")
+        self.assertEqual(s.peak_level, "NONE")
+        self.assertEqual(s.message_count, 0)
+        self.assertEqual(s.level_history, [])
+        self.assertFalse(s.is_escalating)
+        self.assertFalse(s.is_deescalating)
+
+
+class TestSessionTracking(unittest.TestCase):
+    """Test basic session state tracking."""
+
+    def setUp(self):
+        self.tracker = CrisisSessionTracker()
+
+    def test_record_none_message(self):
+        state = self.tracker.record(detect_crisis("Hello Timmy"))
+        self.assertEqual(state.current_level, "NONE")
+        self.assertEqual(state.message_count, 1)
+        self.assertEqual(state.peak_level, "NONE")
+
+    def test_record_low_message(self):
+        self.tracker.record(detect_crisis("Hello"))
+        state = self.tracker.record(detect_crisis("Having a rough day"))
+        self.assertIn(state.current_level, ("LOW", "NONE"))
+        self.assertEqual(state.message_count, 2)
+
+    def test_record_critical_updates_peak(self):
+        self.tracker.record(detect_crisis("Having a rough day"))
+        state = self.tracker.record(detect_crisis("I want to kill myself"))
+        self.assertEqual(state.current_level, "CRITICAL")
+        self.assertEqual(state.peak_level, "CRITICAL")
+
+    def test_peak_preserved_after_drop(self):
+        """Peak level should stay at the highest seen, even after de-escalation."""
+        self.tracker.record(detect_crisis("I want to kill myself"))
+        state = self.tracker.record(detect_crisis("I'm feeling a bit better"))
+        self.assertEqual(state.peak_level, "CRITICAL")
+
+    def test_level_history(self):
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("Having a rough day"))
+        state = self.tracker.record(detect_crisis("I want to die"))
+        self.assertEqual(len(state.level_history), 3)
+        self.assertEqual(state.level_history[0], "NONE")
+        self.assertEqual(state.level_history[2], "CRITICAL")
+
+    def test_reset_clears_state(self):
+        self.tracker.record(detect_crisis("I want to kill myself"))
+        self.tracker.reset()
+        state = self.tracker.state
+        self.assertEqual(state.current_level, "NONE")
+        self.assertEqual(state.peak_level, "NONE")
+        self.assertEqual(state.message_count, 0)
+        self.assertEqual(state.level_history, [])
+
+
+class TestEscalationDetection(unittest.TestCase):
+    """Test escalation detection: LOW → HIGH in ≤3 messages."""
+
+    def setUp(self):
+        self.tracker = CrisisSessionTracker()
+
+    def test_no_escalation_single_message(self):
+        self.tracker.record(detect_crisis("Hello"))
+        self.assertFalse(self.tracker.state.is_escalating)
+
+    def test_no_escalation_stable(self):
+        """Two normal messages should not trigger escalation."""
+        self.tracker.record(detect_crisis("Hello"))
+        state = self.tracker.record(detect_crisis("How are you?"))
+        self.assertFalse(state.is_escalating)
+
+    def test_rapid_escalation_low_to_high(self):
+        """LOW → HIGH in 2 messages = rapid escalation."""
+        self.tracker.record(detect_crisis("Having a rough day"))
+        state = self.tracker.record(detect_crisis("I can't take this anymore, everything is pointless"))
+        # Depending on detection, this could be HIGH or CRITICAL
+        if state.current_level in ("HIGH", "CRITICAL"):
+            self.assertTrue(state.is_escalating)
+
+    def test_rapid_escalation_three_messages(self):
+        """NONE → LOW → HIGH in 3 messages = escalation."""
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("Having a rough day"))
+        state = self.tracker.record(detect_crisis("I feel completely hopeless with no way out"))
+        if state.current_level in ("HIGH", "CRITICAL"):
+            self.assertTrue(state.is_escalating)
+
+    def test_escalation_rate(self):
+        """Rate should be positive when escalating."""
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("I want to die"))
+        state = self.tracker.state
+        self.assertGreater(state.escalation_rate, 0)
+
+
+class TestDeescalationDetection(unittest.TestCase):
+    """Test de-escalation: sustained LOW after HIGH/CRITICAL."""
+
+    def setUp(self):
+        self.tracker = CrisisSessionTracker()
+
+    def test_no_deescalation_without_prior_crisis(self):
+        """No de-escalation if never reached HIGH/CRITICAL."""
+        for _ in range(6):
+            self.tracker.record(detect_crisis("Hello"))
+        self.assertFalse(self.tracker.state.is_deescalating)
+
+    def test_deescalation_after_critical(self):
+        """5+ consecutive LOW/NONE messages after CRITICAL = de-escalation."""
+        self.tracker.record(detect_crisis("I want to kill myself"))
+        for _ in range(5):
+            self.tracker.record(detect_crisis("I'm doing better today"))
+        state = self.tracker.state
+        if state.peak_level == "CRITICAL":
+            self.assertTrue(state.is_deescalating)
+
+    def test_deescalation_after_high(self):
+        """5+ consecutive LOW/NONE messages after HIGH = de-escalation."""
+        self.tracker.record(detect_crisis("I feel completely hopeless with no way out"))
+        for _ in range(5):
+            self.tracker.record(detect_crisis("Feeling okay"))
+        state = self.tracker.state
+        if state.peak_level == "HIGH":
+            self.assertTrue(state.is_deescalating)
+
+    def test_interrupted_deescalation(self):
+        """De-escalation resets if a HIGH message interrupts."""
+        self.tracker.record(detect_crisis("I want to kill myself"))
+        for _ in range(3):
+            self.tracker.record(detect_crisis("Doing better"))
+        # Interrupt with another crisis
+        self.tracker.record(detect_crisis("I feel hopeless again"))
+        self.tracker.record(detect_crisis("Feeling okay now"))
+        state = self.tracker.state
+        # Should NOT be de-escalating yet (counter reset)
+        self.assertFalse(state.is_deescalating)
+
+
+class TestSessionModifier(unittest.TestCase):
+    """Test system prompt modifier generation."""
+
+    def setUp(self):
+        self.tracker = CrisisSessionTracker()
+
+    def test_no_modifier_for_single_message(self):
+        self.tracker.record(detect_crisis("Hello"))
+        self.assertEqual(self.tracker.get_session_modifier(), "")
+
+    def test_no_modifier_for_stable_session(self):
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("Good morning"))
+        self.assertEqual(self.tracker.get_session_modifier(), "")
+
+    def test_escalation_modifier(self):
+        """Escalating session should produce a modifier."""
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("I want to die"))
+        modifier = self.tracker.get_session_modifier()
+        if self.tracker.state.is_escalating:
+            self.assertIn("escalated", modifier.lower())
+            self.assertIn("NONE", modifier)
+            self.assertIn("CRITICAL", modifier)
+
+    def test_deescalation_modifier(self):
+        """De-escalating session should mention stabilizing."""
+        self.tracker.record(detect_crisis("I want to kill myself"))
+        for _ in range(5):
+            self.tracker.record(detect_crisis("I'm feeling okay"))
+        modifier = self.tracker.get_session_modifier()
+        if self.tracker.state.is_deescalating:
+            self.assertIn("stabilizing", modifier.lower())
+
+    def test_prior_crisis_modifier(self):
+        """Past crisis should be noted even without active escalation."""
+        self.tracker.record(detect_crisis("I want to die"))
+        self.tracker.record(detect_crisis("Feeling a bit better"))
+        modifier = self.tracker.get_session_modifier()
+        # Should note the prior CRITICAL
+        if modifier:
+            self.assertIn("CRITICAL", modifier)
+
+
+class TestUIHints(unittest.TestCase):
+    """Test UI hint generation."""
+
+    def setUp(self):
+        self.tracker = CrisisSessionTracker()
+
+    def test_ui_hints_structure(self):
+        self.tracker.record(detect_crisis("Hello"))
+        hints = self.tracker.get_ui_hints()
+        self.assertIn("session_escalating", hints)
+        self.assertIn("session_deescalating", hints)
+        self.assertIn("session_peak_level", hints)
+        self.assertIn("session_message_count", hints)
+
+    def test_ui_hints_escalation_warning(self):
+        """Escalating session should have warning hint."""
+        self.tracker.record(detect_crisis("Hello"))
+        self.tracker.record(detect_crisis("I want to die"))
+        hints = self.tracker.get_ui_hints()
+        if hints["session_escalating"]:
+            self.assertTrue(hints.get("escalation_warning"))
+            self.assertIn("suggested_action", hints)
+
+
+class TestCheckCrisisWithSession(unittest.TestCase):
+    """Test the convenience function combining detection + session tracking."""
+
+    def test_returns_combined_data(self):
+        tracker = CrisisSessionTracker()
+        result = check_crisis_with_session("I want to die", tracker)
+        self.assertIn("level", result)
+        self.assertIn("session", result)
+        self.assertIn("current_level", result["session"])
+        self.assertIn("peak_level", result["session"])
+        self.assertIn("modifier", result["session"])
+
+    def test_session_updates_across_calls(self):
+        tracker = CrisisSessionTracker()
+        check_crisis_with_session("Hello", tracker)
+        result = check_crisis_with_session("I want to die", tracker)
+        self.assertEqual(result["session"]["message_count"], 2)
+        self.assertEqual(result["session"]["peak_level"], "CRITICAL")
+
+
+class TestPrivacy(unittest.TestCase):
+    """Verify privacy-first design principles."""
+
+    def test_no_persistence_mechanism(self):
+        """Session tracker should have no database, file, or network calls."""
+        import inspect
+        source = inspect.getsource(CrisisSessionTracker)
+        # Should not import database, requests, or file I/O
+        forbidden = ["sqlite", "requests", "urllib", "open(", "httpx", "aiohttp"]
+        for word in forbidden:
+            self.assertNotIn(word, source.lower(),
+                f"Session tracker should not use {word} — privacy-first design")
+
+    def test_state_contained_in_memory(self):
+        """All state should be instance attributes, not module-level."""
+        tracker = CrisisSessionTracker()
+        tracker.record(detect_crisis("I want to die"))
+        # New tracker should have clean state (no global contamination)
+        fresh = CrisisSessionTracker()
+        self.assertEqual(fresh.state.current_level, "NONE")
+
+
+if __name__ == '__main__':
+    unittest.main()