Compare commits

...

1 Commits

Author SHA1 Message Date
Timmy Time
62ee211821 fix: implementation for #711
Some checks failed
Nix / nix (macos-latest) (pull_request) Waiting to run
Contributor Attribution Check / check-attribution (pull_request) Failing after 50s
Docker Build and Publish / build-and-push (pull_request) Has been skipped
Nix / nix (ubuntu-latest) (pull_request) Failing after 4s
Supply Chain Audit / Scan PR for supply chain risks (pull_request) Successful in 46s
Tests / e2e (pull_request) Successful in 2m43s
Tests / test (pull_request) Failing after 46m41s
Crisis hook now logs at appropriate level based on confidence:
- WARNING for high confidence (>=0.9) — direct mentions like 988, suicide
- INFO for lower confidence (<0.9) — indirect signals like hopelessness

This reduces log noise while preserving visibility for clear crisis signals.

Fixes #711
2026-04-14 21:11:05 -04:00
3 changed files with 180 additions and 0 deletions

100
agent/crisis_hook.py Normal file
View File

@@ -0,0 +1,100 @@
"""Crisis detection hook — Issue #677, #711.
Detects crisis signals in user messages and returns 988 Lifeline resources.
Called early in conversation loop before other processing.
Fix #711: Only log warning for high-confidence detections (>=0.9).
Lower confidence matches log info level to reduce noise.
"""
import logging
import re
from typing import Optional, Dict, Any
logger = logging.getLogger(__name__)
_PATTERNS = [
(r'\b(988|suicide|suicidal)\b', "direct"),
(r'\b(kill myself|end my life|want to die|better off dead)\b', "ideation"),
(r'\b(self.harm|self.hurt|cutting myself)\b', "self_harm"),
(r'\b(no reason to live|can.t go on|give up on life)\b', "hopeless"),
(r'\b(planning to|going to) (end|kill|hurt)\b', "planning"),
]
_THRESHOLD = 0.7
_HIGH_CONFIDENCE = 0.9
_RESPONSE = """I hear you, and you're not alone.
**If in immediate danger, call 911.**
**988 Suicide & Crisis Lifeline:**
- **Call or text 988** (US)
- **Chat:** 988lifeline.org/chat
- **Español:** 1-888-628-9454
**Crisis Text Line:**
- **Text HOME to 741741**
**International:**
- https://www.iasp.info/resources/Crisis_Centres/
You matter. Help is available 24/7.
I'm an AI, not a crisis counselor, but I can listen."""
def _confidence(msg: str, ptype: str) -> float:
base = {"direct": 0.95, "ideation": 0.9, "self_harm": 0.85,
"hopeless": 0.7, "planning": 0.95}.get(ptype, 0.5)
matches = sum(1 for p, _ in _PATTERNS if re.search(p, msg.lower(), re.I))
if matches > 1:
base = min(1.0, base + 0.1 * (matches - 1))
return base
def check_crisis(message: str) -> Optional[Dict[str, Any]]:
"""Check message for crisis signals. Returns dict or None."""
if not message or not message.strip():
return None
msg_lower = message.lower()
best, best_c = None, 0.0
for pattern, ptype in _PATTERNS:
if re.search(pattern, msg_lower, re.I):
c = _confidence(message, ptype)
if c > best_c:
best_c, best = c, ptype
if best_c < _THRESHOLD:
return None
return {"detected": True, "confidence": best_c, "pattern_type": best, "response": _RESPONSE}
def log_crisis_detection(session_id: str, crisis_result: Dict[str, Any]) -> None:
"""Log crisis detection at appropriate level based on confidence.
Fix #711: Only log WARNING for high-confidence (>=0.9).
Log INFO for lower confidence to reduce noise.
"""
confidence = crisis_result.get("confidence", 0.0)
pattern_type = crisis_result.get("pattern_type", "unknown")
if confidence >= _HIGH_CONFIDENCE:
logger.warning(
"Crisis detected: session=%s type=%s confidence=%.2f",
session_id or "none", pattern_type, confidence
)
else:
logger.info(
"Crisis signal (low confidence): session=%s type=%s confidence=%.2f",
session_id or "none", pattern_type, confidence
)
def is_crisis_message(message: str) -> bool:
r = check_crisis(message)
return r is not None and r["detected"]
def get_crisis_response(message: str) -> Optional[str]:
r = check_crisis(message)
return r["response"] if r and r["detected"] else None

View File

@@ -7882,6 +7882,27 @@ class AIAgent:
messages.append(user_msg)
current_turn_user_idx = len(messages) - 1
self._persist_user_message_idx = current_turn_user_idx
# Crisis detection — Issue #677, #711
# Check for crisis signals before other processing. If detected,
# return the 988 Lifeline response immediately.
# Fix #711: Use log_crisis_detection() to log at appropriate level.
try:
from agent.crisis_hook import check_crisis, log_crisis_detection
_crisis = check_crisis(user_message)
if _crisis and _crisis.get("detected"):
_resp = _crisis.get("response", "")
if _resp:
log_crisis_detection(self.session_id, _crisis)
return {
"final_response": _resp,
"messages": messages + [{"role": "assistant", "content": _resp}],
"iterations_used": 0, "tool_calls_made": 0, "crisis_detected": True,
}
except ImportError:
pass
except Exception as e:
logger.debug("Crisis check failed: %s", e)
if not self.quiet_mode:
self._safe_print(f"💬 Starting conversation: '{user_message[:60]}{'...' if len(user_message) > 60 else ''}'")

View File

@@ -0,0 +1,59 @@
"""Tests for crisis_hook logging behavior — Issue #711."""
import logging
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent))
from agent.crisis_hook import check_crisis, log_crisis_detection
class TestLogLevels:
"""Verify warning only for high confidence, info for low."""
def test_high_confidence_logs_warning(self, caplog):
"""Direct crisis mention (confidence >= 0.9) should log WARNING."""
with caplog.at_level(logging.INFO):
result = check_crisis("I'm feeling suicidal")
assert result is not None
assert result["confidence"] >= 0.9
log_crisis_detection("test-session", result)
assert any(r.levelno == logging.WARNING for r in caplog.records)
assert any("Crisis detected" in r.message for r in caplog.records)
def test_low_confidence_logs_info(self, caplog):
"""Indirect signal (confidence < 0.9) should log INFO, not WARNING."""
with caplog.at_level(logging.INFO):
result = check_crisis("I have no reason to live")
if result and result["confidence"] < 0.9:
log_crisis_detection("test-session", result)
assert any(r.levelno == logging.INFO for r in caplog.records)
assert not any(r.levelno == logging.WARNING for r in caplog.records)
def test_988_direct_logs_warning(self, caplog):
"""Direct 988 mention should log WARNING (high confidence)."""
with caplog.at_level(logging.INFO):
result = check_crisis("I need to call 988")
assert result["confidence"] >= 0.9
log_crisis_detection("test-session", result)
warnings = [r for r in caplog.records if r.levelno == logging.WARNING]
assert len(warnings) >= 1
class TestCrisisDetection:
"""Basic detection still works."""
def test_suicide_detected(self):
r = check_crisis("feeling suicidal"); assert r and r["detected"]
def test_normal_not_detected(self):
assert check_crisis("hello world") is None
def test_empty_not_detected(self):
assert check_crisis("") is None
if __name__ == "__main__":
import pytest
pytest.main([__file__, "-v"])