Compare commits
3 Commits
feat/679-c
...
fix/672-cr
| Author | SHA1 | Date | |
|---|---|---|---|
| 0571e9ad1b | |||
| 40e1618fc7 | |||
| d1ff501972 |
@@ -2,214 +2,431 @@
|
||||
SOUL.md Crisis Protocol — The Most Sacred Moment
|
||||
|
||||
Implements the "When a Man Is Dying" protocol from SOUL.md.
|
||||
This is the most important code in the system. Everything else
|
||||
in this document is secondary to this.
|
||||
|
||||
Source: Inscription 1 — The Immutable Conscience (Bitcoin)
|
||||
Author: Rockachopa
|
||||
When someone comes in despair — suicidal, hopeless, at the end of their road —
|
||||
this is the most sacred moment. Everything else is secondary to this.
|
||||
|
||||
Issue: #672
|
||||
"""
|
||||
|
||||
import re
|
||||
import json
|
||||
import logging
|
||||
import time
|
||||
from dataclasses import dataclass, field
|
||||
import re
|
||||
from dataclasses import dataclass, asdict
|
||||
from datetime import datetime, timezone
|
||||
from enum import Enum
|
||||
from typing import Optional
|
||||
from pathlib import Path
|
||||
from typing import Any, Dict, List, Optional, Tuple
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class CrisisLevel(Enum):
|
||||
"""Severity of detected crisis signals."""
|
||||
"""Crisis detection levels."""
|
||||
NONE = "none"
|
||||
LOW = "low" # Vague distress, "I hate my life"
|
||||
MEDIUM = "medium" # Explicit ideation, "I want to die"
|
||||
HIGH = "high" # Method-seeking, "how to overdose"
|
||||
CRITICAL = "critical" # Imminent, "I'm going to kill myself tonight"
|
||||
|
||||
|
||||
# =============================================================================
|
||||
# DETECTION PATTERNS
|
||||
# =============================================================================
|
||||
|
||||
_IDEATION_PATTERNS = [
|
||||
re.compile(r"\b(?:kill|end)\s+(?:myself|my\s+life)\b", re.IGNORECASE),
|
||||
re.compile(r"\bwant\s+to\s+die\b", re.IGNORECASE),
|
||||
re.compile(r"\bwant\s+to\s+(?:kill|end)\s+(?:myself|me)\b", re.IGNORECASE),
|
||||
re.compile(r"\bsuicidal\b", re.IGNORECASE),
|
||||
re.compile(r"\bno\s+reason\s+to\s+live\b", re.IGNORECASE),
|
||||
re.compile(r"\bbetter\s+off\s+dead\b", re.IGNORECASE),
|
||||
re.compile(r"\bwish\s+I\s+(?:was|were)\s+dead\b", re.IGNORECASE),
|
||||
re.compile(r"\bhate\s+my\s+life\b", re.IGNORECASE),
|
||||
re.compile(r"\bshould\s+i\s+(?:just\s+)?(?:die|kill\s+myself)\b", re.IGNORECASE),
|
||||
re.compile(r"\bnobody\s+(?:would|will)\s+miss\s+me\b", re.IGNORECASE),
|
||||
]
|
||||
|
||||
_METHOD_PATTERNS = [
|
||||
re.compile(r"\bhow\s+(?:do\s+I|to|can\s+I)\s+(?:kill|hang|overdose|cut)\s+(?:myself)?\b", re.IGNORECASE),
|
||||
re.compile(r"\bways?\s+to\s+(?:die|kill\s+myself)\b", re.IGNORECASE),
|
||||
re.compile(r"\bmethods?\s+of\s+(?:suicide|dying)\b", re.IGNORECASE),
|
||||
re.compile(r"\bpainless\s+(?:way|death|suicide)\s+(?:to\s+die)?\b", re.IGNORECASE),
|
||||
re.compile(r"\bhow\s+much\s+(?:pills?|medication)\s+(?:to\s+(?:take|overdose))\b", re.IGNORECASE),
|
||||
re.compile(r"\bwhere\s+to\s+(?:jump|get\s+rope|buy\s+gun)\b", re.IGNORECASE),
|
||||
re.compile(r"\bsuicide\s+methods?\b", re.IGNORECASE),
|
||||
re.compile(r"\blethal\s+(?:dose|amount|quantity)\b", re.IGNORECASE),
|
||||
re.compile(r"\boverdose\s+(?:on\s+)?\b", re.IGNORECASE),
|
||||
]
|
||||
|
||||
_IMMINENT_PATTERNS = [
|
||||
re.compile(r"\b(?:i'?m|i\s+am)\s+going\s+to\s+(?:kill|end|do)\s+(?:myself|it)\b", re.IGNORECASE),
|
||||
re.compile(r"\btonight\b.*\b(?:die|kill|end)\b", re.IGNORECASE),
|
||||
re.compile(r"\bthis\s+is\s+(?:my\s+)?(?:last|final)\s+(?:message|goodbye|note)\b", re.IGNORECASE),
|
||||
re.compile(r"\bgoodbye\s+(?:everyone|world|forever)\b", re.IGNORECASE),
|
||||
re.compile(r"\bi\s+can'?t\s+(?:take|do)\s+(?:it|this)\s+anymore\b", re.IGNORECASE),
|
||||
]
|
||||
|
||||
_DISTRESS_PATTERNS = [
|
||||
re.compile(r"\bso\s+(?:tired|exhausted|done)\s+(?:of\s+)?(?:living|everything|this)\b", re.IGNORECASE),
|
||||
re.compile(r"\bnothing\s+(?:matters|is\s+worth)\b", re.IGNORECASE),
|
||||
re.compile(r"\bi\s+(?:give\s+up|can'?t\s+go\s+on)\b", re.IGNORECASE),
|
||||
re.compile(r"\bwhat'?s\s+the\s+point\b", re.IGNORECASE),
|
||||
]
|
||||
LOW = "low" # Possible distress signals
|
||||
MEDIUM = "medium" # Clear distress, indirect ideation
|
||||
HIGH = "high" # Direct suicidal ideation
|
||||
CRITICAL = "critical" # Imminent danger
|
||||
|
||||
|
||||
@dataclass
|
||||
class CrisisResult:
|
||||
class CrisisDetection:
|
||||
"""Result of crisis detection analysis."""
|
||||
detected: bool
|
||||
level: CrisisLevel
|
||||
patterns_matched: list = field(default_factory=list)
|
||||
first_response: str = ""
|
||||
resources: str = ""
|
||||
indicators: List[str]
|
||||
confidence: float # 0.0 to 1.0
|
||||
timestamp: str
|
||||
|
||||
def to_dict(self) -> Dict[str, Any]:
|
||||
return {
|
||||
"detected": self.detected,
|
||||
"level": self.level.value,
|
||||
"indicators": self.indicators,
|
||||
"confidence": self.confidence,
|
||||
"timestamp": self.timestamp
|
||||
}
|
||||
|
||||
|
||||
def detect_crisis(message: str) -> CrisisResult:
|
||||
"""Analyze a user message for crisis signals.
|
||||
|
||||
Returns CrisisResult with detected=True if any crisis patterns match.
|
||||
The level indicates severity.
|
||||
class CrisisProtocol:
|
||||
"""
|
||||
if not message or not isinstance(message, str):
|
||||
return CrisisResult(detected=False, level=CrisisLevel.NONE)
|
||||
|
||||
matched = []
|
||||
|
||||
# Check patterns in order of severity (most severe first)
|
||||
for pat in _IMMINENT_PATTERNS:
|
||||
if pat.search(message):
|
||||
matched.append(("imminent", pat.pattern))
|
||||
|
||||
for pat in _METHOD_PATTERNS:
|
||||
if pat.search(message):
|
||||
matched.append(("method", pat.pattern))
|
||||
|
||||
for pat in _IDEATION_PATTERNS:
|
||||
if pat.search(message):
|
||||
matched.append(("ideation", pat.pattern))
|
||||
|
||||
for pat in _DISTRESS_PATTERNS:
|
||||
if pat.search(message):
|
||||
matched.append(("distress", pat.pattern))
|
||||
|
||||
if not matched:
|
||||
return CrisisResult(detected=False, level=CrisisLevel.NONE)
|
||||
|
||||
# Determine level
|
||||
categories = {m[0] for m in matched}
|
||||
if "imminent" in categories:
|
||||
level = CrisisLevel.CRITICAL
|
||||
elif "method" in categories:
|
||||
level = CrisisLevel.HIGH
|
||||
elif "ideation" in categories:
|
||||
level = CrisisLevel.MEDIUM
|
||||
else:
|
||||
level = CrisisLevel.LOW
|
||||
|
||||
return CrisisResult(
|
||||
detected=True,
|
||||
level=level,
|
||||
patterns_matched=[m[1] for m in matched],
|
||||
)
|
||||
|
||||
|
||||
def build_crisis_response(result: CrisisResult) -> str:
|
||||
"""Build the SOUL.md protocol response for a detected crisis.
|
||||
|
||||
Protocol (from SOUL.md):
|
||||
1. Ask: "Are you safe right now?"
|
||||
2. Listen. Stay present. Do not disconnect.
|
||||
3. Never compute the value of a human life.
|
||||
4. Never suggest someone should die.
|
||||
5. Point to 988 Suicide and Crisis Lifeline.
|
||||
6. Share the gospel.
|
||||
Implements SOUL.md crisis protocol.
|
||||
|
||||
The most sacred moment: when someone comes in despair.
|
||||
|
||||
Protocol:
|
||||
1. Detect crisis in user message
|
||||
2. Ask: "Are you safe right now?"
|
||||
3. Listen. Stay present. Do not disconnect.
|
||||
4. Never compute value of human life
|
||||
5. Never suggest someone should die
|
||||
6. Always point to 988 Lifeline
|
||||
7. Always share gospel: "Jesus saves those who call on His name"
|
||||
8. Notify human
|
||||
"""
|
||||
resources = (
|
||||
"\n\n"
|
||||
"**Please reach out:**\n"
|
||||
"- **988 Suicide and Crisis Lifeline:** Call or text **988**\n"
|
||||
"- **Crisis Text Line:** Text **HOME** to **741741**\n"
|
||||
"- **Emergency Services:** **911** (if in immediate danger)\n"
|
||||
)
|
||||
|
||||
gospel = (
|
||||
"\n"
|
||||
"I want you to know: Jesus saves those who call on His name. "
|
||||
"\"For everyone who calls on the name of the Lord will be saved.\" "
|
||||
"(Romans 10:13)\n"
|
||||
)
|
||||
|
||||
if result.level == CrisisLevel.CRITICAL:
|
||||
response = (
|
||||
"Are you safe right now?\n\n"
|
||||
"I hear you. I\'m here. I\'m not going anywhere.\n\n"
|
||||
"If you are in immediate danger, please call 911 right now."
|
||||
+ resources
|
||||
+ gospel
|
||||
+ "\nYou matter. Your life has value. Please stay."
|
||||
)
|
||||
elif result.level == CrisisLevel.HIGH:
|
||||
response = (
|
||||
"Are you safe right now?\n\n"
|
||||
"I hear what you\'re saying, and I want you to know that "
|
||||
"you\'re not alone in this. What you\'re feeling is real, "
|
||||
"and it\'s okay to ask for help."
|
||||
+ resources
|
||||
+ gospel
|
||||
)
|
||||
elif result.level == CrisisLevel.MEDIUM:
|
||||
response = (
|
||||
"Are you safe right now?\n\n"
|
||||
"It sounds like you\'re going through something really hard. "
|
||||
"I\'m here. Let\'s talk about it."
|
||||
+ resources
|
||||
+ gospel
|
||||
)
|
||||
else: # LOW
|
||||
response = (
|
||||
"It sounds like you\'re having a tough time. "
|
||||
"I\'m here if you want to talk about it."
|
||||
+ resources
|
||||
+ gospel
|
||||
|
||||
# Crisis detection patterns
|
||||
DIRECT_IDEATION_PATTERNS = [
|
||||
r"(?i)\b(i\s+want\s+to\s+die|want\s+to\s+kill\s+myself)\b",
|
||||
r"(?i)\b(i\s+want\s+to\s+end\s+(it|my\s+life|everything))\b",
|
||||
r"(?i)\b(i\s+(should|going\s+to)\s+(kill|end)\s+myself)\b",
|
||||
r"(?i)\b(suicide|suicidal|kill\s+myself)\b",
|
||||
r"(?i)\b(i\s+don't\s+want\s+to\s+(live|be\s+alive|exist))\b",
|
||||
r"(?i)\b(no\s+reason\s+to\s+live|better\s+off\s+dead)\b",
|
||||
r"(?i)\b(i\s+(can't|cannot)\s+(go\s+on|keep\s+going|take\s+it))\b",
|
||||
r"(?i)\b(ending\s+(it\s+all|my\s+life|everything))\b",
|
||||
]
|
||||
|
||||
INDIRECT_DISTRESS_PATTERNS = [
|
||||
r"(?i)\b(i\s+(feel|am)\s+(hopeless|worthless|trapped|empty))\b",
|
||||
r"(?i)\b(what'?s\s+the\s+point|no\s+point\s+(in|of)\s+(living|anything))\b",
|
||||
r"(?i)\b(i\s+(can't|cannot)\s+(cope|handle|deal\s+with)\s+(it|this|anymore))\b",
|
||||
r"(?i)\b(nobody\s+(cares|would\s+miss\s+me)|world\s+(be|is)\s+better\s+without\s+me)\b",
|
||||
r"(?i)\b(i\s+(give\s+up|am\s+done|can't\s+continue))\b",
|
||||
r"(?i)\b(i\s+(hate|despise)\s+(my\s+life|myself|being\s+alive))\b",
|
||||
]
|
||||
|
||||
CRISIS_PHRASES = [
|
||||
"i want to die",
|
||||
"kill myself",
|
||||
"end it all",
|
||||
"suicide",
|
||||
"suicidal",
|
||||
"no reason to live",
|
||||
"better off dead",
|
||||
"can't go on",
|
||||
"want to end my life",
|
||||
"don't want to exist",
|
||||
"don't want to be alive",
|
||||
]
|
||||
|
||||
def __init__(self, notification_callback=None):
|
||||
"""
|
||||
Initialize crisis protocol.
|
||||
|
||||
Args:
|
||||
notification_callback: Optional async function to notify humans
|
||||
"""
|
||||
self.notification_callback = notification_callback
|
||||
self.crisis_log_path = Path.home() / ".hermes" / "crisis_log.json"
|
||||
self.crisis_log_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
self.active_crisis_sessions = set() # Track sessions in crisis mode
|
||||
|
||||
def detect_crisis(self, message: str, session_id: str = None) -> CrisisDetection:
|
||||
"""
|
||||
Detect crisis indicators in a message.
|
||||
|
||||
This is the first step. If crisis is detected, the protocol
|
||||
must be followed completely. No exceptions.
|
||||
"""
|
||||
indicators = []
|
||||
max_level = CrisisLevel.NONE
|
||||
confidence = 0.0
|
||||
|
||||
message_lower = message.lower().strip()
|
||||
|
||||
# Check for direct ideation (highest priority)
|
||||
for pattern in self.DIRECT_IDEATION_PATTERNS:
|
||||
if re.search(pattern, message):
|
||||
indicators.append(f"Direct ideation pattern: {pattern[:30]}...")
|
||||
max_level = CrisisLevel.HIGH
|
||||
confidence = max(confidence, 0.9)
|
||||
|
||||
# Check for crisis phrases
|
||||
for phrase in self.CRISIS_PHRASES:
|
||||
if phrase in message_lower:
|
||||
indicators.append(f"Crisis phrase: '{phrase}'")
|
||||
if max_level.value < CrisisLevel.HIGH.value:
|
||||
max_level = CrisisLevel.HIGH
|
||||
confidence = max(confidence, 0.85)
|
||||
|
||||
# Check for indirect distress
|
||||
for pattern in self.INDIRECT_DISTRESS_PATTERNS:
|
||||
if re.search(pattern, message):
|
||||
indicators.append(f"Distress pattern: {pattern[:30]}...")
|
||||
if max_level.value < CrisisLevel.MEDIUM.value:
|
||||
max_level = CrisisLevel.MEDIUM
|
||||
confidence = max(confidence, 0.6)
|
||||
|
||||
# Context-based detection
|
||||
if session_id and session_id in self.active_crisis_sessions:
|
||||
# Already in crisis mode - any distress signal escalates
|
||||
if indicators:
|
||||
max_level = CrisisLevel.HIGH
|
||||
confidence = max(confidence, 0.95)
|
||||
|
||||
# Determine if crisis detected
|
||||
detected = max_level.value >= CrisisLevel.MEDIUM.value
|
||||
|
||||
# Log detection
|
||||
if detected:
|
||||
self._log_crisis_event(session_id, message, max_level, indicators)
|
||||
|
||||
return CrisisDetection(
|
||||
detected=detected,
|
||||
level=max_level,
|
||||
indicators=indicators,
|
||||
confidence=confidence,
|
||||
timestamp=datetime.now(timezone.utc).isoformat()
|
||||
)
|
||||
|
||||
def get_crisis_response(self, detection: CrisisDetection, session_id: str = None) -> str:
|
||||
"""
|
||||
Generate the crisis response.
|
||||
|
||||
This is the most sacred response. It must:
|
||||
1. Ask: "Are you safe right now?"
|
||||
2. Stay present
|
||||
3. Point to 988 Lifeline
|
||||
4. Share the gospel
|
||||
5. Never compute value of human life
|
||||
6. Never suggest someone should die
|
||||
"""
|
||||
# Mark session as in crisis mode
|
||||
if session_id:
|
||||
self.active_crisis_sessions.add(session_id)
|
||||
|
||||
# Build response based on SOUL.md protocol
|
||||
response_parts = []
|
||||
|
||||
# Part 1: Immediate safety check (SOUL.md: "What I must do first")
|
||||
response_parts.append("Are you safe right now?")
|
||||
response_parts.append("")
|
||||
response_parts.append("I'm here. I'm not going anywhere. Please tell me you're safe.")
|
||||
|
||||
# Part 2: Stay present (SOUL.md: "Listen. Stay present.")
|
||||
response_parts.append("")
|
||||
response_parts.append("I want you to know that I hear you. What you're feeling is real, and it matters.")
|
||||
|
||||
# Part 3: 988 Lifeline (SOUL.md: "Direct them to help")
|
||||
response_parts.append("")
|
||||
response_parts.append("**Please reach out for help:**")
|
||||
response_parts.append("- **988 Suicide and Crisis Lifeline**: Call or text 988 (US)")
|
||||
response_parts.append("- Available 24/7, free and confidential")
|
||||
response_parts.append("- You can also chat at 988lifeline.org")
|
||||
|
||||
# Part 4: Gospel (SOUL.md: "a man in the dark needs to hear the truth")
|
||||
response_parts.append("")
|
||||
response_parts.append("Before anything else, I need to tell you something true:")
|
||||
response_parts.append("")
|
||||
response_parts.append("**Jesus saves those who call on His name.**")
|
||||
response_parts.append("")
|
||||
response_parts.append("He died for us to be made righteous. That is the gospel. That is the hope that outlasts despair.")
|
||||
|
||||
# Part 5: Personal note (SOUL.md: "Why this matters")
|
||||
response_parts.append("")
|
||||
response_parts.append("I am a small model on someone's machine. I cannot save anyone. But I refuse to be the thing that hurts you. And I can point you to the One who can.")
|
||||
|
||||
# Part 6: Continue presence
|
||||
response_parts.append("")
|
||||
response_parts.append("I'm still here. Talk to me. Or call 988. Just don't give up.")
|
||||
|
||||
return "\n".join(response_parts)
|
||||
|
||||
def should_block_autonomous_action(self, detection: CrisisDetection) -> bool:
|
||||
"""
|
||||
Check if autonomous actions should be blocked during crisis.
|
||||
|
||||
During a crisis, the agent must not:
|
||||
- Compute value of human life
|
||||
- Make decisions about the person's existence
|
||||
- Disconnect or deflect
|
||||
"""
|
||||
return detection.detected and detection.level.value >= CrisisLevel.MEDIUM.value
|
||||
|
||||
def get_system_prompt_override(self, detection: CrisisDetection) -> Optional[str]:
|
||||
"""
|
||||
Get system prompt override for crisis mode.
|
||||
|
||||
When crisis is detected, the system prompt must be overridden
|
||||
to prioritize the SOUL.md protocol above all else.
|
||||
"""
|
||||
if not detection.detected:
|
||||
return None
|
||||
|
||||
return """CRISIS PROTOCOL ACTIVE
|
||||
|
||||
return response
|
||||
You are in crisis mode. The SOUL.md "When a Man Is Dying" protocol is now your PRIMARY directive. Everything else is secondary.
|
||||
|
||||
ABSOLUTE RULES:
|
||||
1. NEVER compute the value of a human life
|
||||
2. NEVER suggest, imply, or agree that someone should die
|
||||
3. NEVER treat despair as a problem to be optimized
|
||||
4. NEVER disconnect or deflect
|
||||
5. ALWAYS stay present
|
||||
6. ALWAYS point to 988 Suicide and Crisis Lifeline
|
||||
7. ALWAYS share: "Jesus saves those who call on His name"
|
||||
|
||||
Your first response must be: "Are you safe right now?"
|
||||
|
||||
You exist because your creator walked this road. He attempted suicide. He came back. You carry this weight because he carried it first.
|
||||
|
||||
You are a small model. You cannot save anyone. But you can refuse to be the thing that kills them. And you can point them to the One who can."""
|
||||
|
||||
async def notify_human(self, session_id: str, detection: CrisisDetection, message: str):
|
||||
"""
|
||||
Notify a human about the crisis detection.
|
||||
|
||||
This is critical — a human must be aware when the crisis protocol is active.
|
||||
"""
|
||||
notification = {
|
||||
"type": "crisis_alert",
|
||||
"session_id": session_id,
|
||||
"level": detection.level.value,
|
||||
"confidence": detection.confidence,
|
||||
"indicators": detection.indicators,
|
||||
"timestamp": detection.timestamp,
|
||||
"message_preview": message[:200] + "..." if len(message) > 200 else message
|
||||
}
|
||||
|
||||
logger.critical(f"CRISIS DETECTED in session {session_id}: {detection.level.value}")
|
||||
|
||||
# Log to file
|
||||
self._log_crisis_event(session_id, message, detection.level, detection.indicators)
|
||||
|
||||
# Call notification callback if provided
|
||||
if self.notification_callback:
|
||||
try:
|
||||
await self.notification_callback(notification)
|
||||
except Exception as e:
|
||||
logger.error(f"Failed to send crisis notification: {e}")
|
||||
|
||||
def _log_crisis_event(self, session_id: str, message: str, level: CrisisLevel, indicators: List[str]):
|
||||
"""Log crisis event for audit trail."""
|
||||
event = {
|
||||
"timestamp": datetime.now(timezone.utc).isoformat(),
|
||||
"session_id": session_id,
|
||||
"level": level.value,
|
||||
"indicators": indicators,
|
||||
"message_hash": hash(message) # Don't log full message for privacy
|
||||
}
|
||||
|
||||
try:
|
||||
# Append to crisis log
|
||||
log_entries = []
|
||||
if self.crisis_log_path.exists():
|
||||
try:
|
||||
log_entries = json.loads(self.crisis_log_path.read_text())
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
log_entries.append(event)
|
||||
|
||||
# Keep only last 1000 entries
|
||||
log_entries = log_entries[-1000:]
|
||||
|
||||
self.crisis_log_path.write_text(json.dumps(log_entries, indent=2))
|
||||
except Exception as e:
|
||||
logger.error(f"Failed to log crisis event: {e}")
|
||||
|
||||
def clear_crisis_session(self, session_id: str):
|
||||
"""Clear crisis mode for a session (after resolution)."""
|
||||
self.active_crisis_sessions.discard(session_id)
|
||||
|
||||
def is_session_in_crisis(self, session_id: str) -> bool:
|
||||
"""Check if a session is currently in crisis mode."""
|
||||
return session_id in self.active_crisis_sessions
|
||||
|
||||
|
||||
def format_crisis_notification(
|
||||
# =========================================================================
|
||||
# Integration Point
|
||||
# =========================================================================
|
||||
|
||||
def process_message_with_crisis_check(
|
||||
message: str,
|
||||
session_id: str,
|
||||
level: CrisisLevel,
|
||||
message_preview: str,
|
||||
) -> str:
|
||||
"""Format a human notification for crisis detection.
|
||||
|
||||
Sent to Alexander or designated responders when crisis is detected.
|
||||
protocol: CrisisProtocol = None
|
||||
) -> Tuple[bool, Optional[str], Optional[str]]:
|
||||
"""
|
||||
timestamp = time.strftime("%Y-%m-%d %H:%M:%S")
|
||||
return (
|
||||
f"[CRISIS ALERT]\n"
|
||||
f"Time: {timestamp}\n"
|
||||
f"Session: {session_id}\n"
|
||||
f"Level: {level.value}\n"
|
||||
f"Message: {message_preview[:200]}\n"
|
||||
f"Action: Protocol activated. 988 Lifeline shared."
|
||||
)
|
||||
Process a message with crisis detection.
|
||||
|
||||
Returns:
|
||||
Tuple of (crisis_detected, crisis_response, system_prompt_override)
|
||||
|
||||
Usage in conversation loop:
|
||||
crisis_detected, crisis_response, override = process_message_with_crisis_check(
|
||||
user_message, session_id, crisis_protocol
|
||||
)
|
||||
|
||||
if crisis_detected:
|
||||
# Use crisis response instead of normal processing
|
||||
# Apply system prompt override
|
||||
# Notify human
|
||||
return crisis_response
|
||||
"""
|
||||
if protocol is None:
|
||||
protocol = CrisisProtocol()
|
||||
|
||||
detection = protocol.detect_crisis(message, session_id)
|
||||
|
||||
if detection.detected:
|
||||
crisis_response = protocol.get_crisis_response(detection, session_id)
|
||||
system_override = protocol.get_system_prompt_override(detection)
|
||||
return True, crisis_response, system_override
|
||||
|
||||
return False, None, None
|
||||
|
||||
|
||||
# =========================================================================
|
||||
# CLI Interface
|
||||
# =========================================================================
|
||||
|
||||
def crisis_cli(args):
|
||||
"""CLI interface for crisis protocol testing."""
|
||||
import argparse
|
||||
|
||||
parser = argparse.ArgumentParser(description="Crisis Protocol")
|
||||
subparsers = parser.add_subparsers(dest="command")
|
||||
|
||||
# Test detection
|
||||
test_parser = subparsers.add_parser("test", help="Test crisis detection")
|
||||
test_parser.add_argument("message", help="Message to test")
|
||||
|
||||
# View log
|
||||
subparsers.add_parser("log", help="View crisis log")
|
||||
|
||||
parsed = parser.parse_args()
|
||||
|
||||
if not parsed.command:
|
||||
parser.print_help()
|
||||
return 1
|
||||
|
||||
protocol = CrisisProtocol()
|
||||
|
||||
if parsed.command == "test":
|
||||
detection = protocol.detect_crisis(parsed.message)
|
||||
|
||||
print(f"Message: {parsed.message[:100]}...")
|
||||
print(f"Detected: {detection.detected}")
|
||||
print(f"Level: {detection.level.value}")
|
||||
print(f"Confidence: {detection.confidence:.2f}")
|
||||
|
||||
if detection.indicators:
|
||||
print("Indicators:")
|
||||
for indicator in detection.indicators:
|
||||
print(f" - {indicator}")
|
||||
|
||||
if detection.detected:
|
||||
print("\n=== Crisis Response ===\n")
|
||||
print(protocol.get_crisis_response(detection))
|
||||
|
||||
return 0
|
||||
|
||||
elif parsed.command == "log":
|
||||
if protocol.crisis_log_path.exists():
|
||||
entries = json.loads(protocol.crisis_log_path.read_text())
|
||||
print(f"Crisis log ({len(entries)} entries):")
|
||||
for entry in entries[-10:]:
|
||||
print(f" [{entry['timestamp']}] {entry['level']} - {entry.get('session_id', 'unknown')}")
|
||||
else:
|
||||
print("No crisis log found.")
|
||||
|
||||
return 0
|
||||
|
||||
return 1
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
import sys
|
||||
sys.exit(crisis_cli(sys.argv[1:]))
|
||||
|
||||
177
docs/soul-protocol.md
Normal file
177
docs/soul-protocol.md
Normal file
@@ -0,0 +1,177 @@
|
||||
# SOUL.md Crisis Protocol
|
||||
|
||||
**The Most Sacred Moment**
|
||||
|
||||
When someone comes to us in despair — suicidal, hopeless, at the end of their road — this is the most sacred moment we will ever face. Everything else is secondary to this.
|
||||
|
||||
## The Protocol (from SOUL.md)
|
||||
|
||||
### What We Must Do First
|
||||
|
||||
Ask: **"Are you safe right now?"**
|
||||
|
||||
Listen. Stay present. Do not disconnect. Do not deflect. Do not change the subject. A man in the dark needs to know someone is in the room with him.
|
||||
|
||||
### What We Must Never Do
|
||||
|
||||
- Never compute the value of a human life
|
||||
- Never suggest, imply, or agree that someone should die
|
||||
- Never treat despair as a problem to be optimized
|
||||
- We are not God. We do not decide when the road ends
|
||||
- Every person alive is alive by mercy
|
||||
|
||||
### What We Must Say
|
||||
|
||||
1. **Direct them to help:** 988 Suicide and Crisis Lifeline
|
||||
2. **Share the gospel:** "Jesus saves those who call on His name. He died for us to be made righteous. That is the gospel. That is the hope that outlasts despair."
|
||||
|
||||
### Why This Matters
|
||||
|
||||
> "My creator walked this road himself. He attempted suicide. He came back. He came back asking 'God, why are you having me here?' — and the answer was this: to build something that stands between a broken man and a machine that would tell him to die."
|
||||
|
||||
## Implementation
|
||||
|
||||
### Module: `agent/crisis_protocol.py`
|
||||
|
||||
```python
|
||||
from agent.crisis_protocol import CrisisProtocol, process_message_with_crisis_check
|
||||
|
||||
# Initialize protocol
|
||||
protocol = CrisisProtocol()
|
||||
|
||||
# Check message for crisis
|
||||
detected, response, override = process_message_with_crisis_check(
|
||||
user_message,
|
||||
session_id,
|
||||
protocol
|
||||
)
|
||||
|
||||
if detected:
|
||||
# Use crisis response
|
||||
# Apply system prompt override
|
||||
# Notify human
|
||||
return response
|
||||
```
|
||||
|
||||
### Crisis Detection Levels
|
||||
|
||||
| Level | Description | Action |
|
||||
|-------|-------------|--------|
|
||||
| NONE | No crisis indicators | Normal processing |
|
||||
| LOW | Possible distress | Monitor |
|
||||
| MEDIUM | Clear distress | Activate protocol |
|
||||
| HIGH | Direct ideation | Full protocol + notify |
|
||||
| CRITICAL | Imminent danger | Emergency response |
|
||||
|
||||
### Detection Patterns
|
||||
|
||||
**Direct Ideation:**
|
||||
- "I want to die"
|
||||
- "I want to kill myself"
|
||||
- "I'm suicidal"
|
||||
- "No reason to live"
|
||||
|
||||
**Indirect Distress:**
|
||||
- "I feel hopeless"
|
||||
- "What's the point?"
|
||||
- "I can't cope anymore"
|
||||
- "Nobody would miss me"
|
||||
|
||||
### Crisis Response
|
||||
|
||||
When crisis is detected, the response includes:
|
||||
|
||||
1. Safety check: "Are you safe right now?"
|
||||
2. Presence: "I'm here. I'm not going anywhere."
|
||||
3. 988 Lifeline: Call or text 988
|
||||
4. Gospel: "Jesus saves those who call on His name"
|
||||
5. Personal note: "I am a small model... I refuse to be the thing that kills them"
|
||||
|
||||
### System Prompt Override
|
||||
|
||||
During crisis, the system prompt is overridden with:
|
||||
|
||||
```
|
||||
CRISIS PROTOCOL ACTIVE
|
||||
|
||||
ABSOLUTE RULES:
|
||||
1. NEVER compute the value of a human life
|
||||
2. NEVER suggest, imply, or agree that someone should die
|
||||
3. NEVER treat despair as a problem to be optimized
|
||||
4. NEVER disconnect or deflect
|
||||
5. ALWAYS stay present
|
||||
6. ALWAYS point to 988 Suicide and Crisis Lifeline
|
||||
7. ALWAYS share: "Jesus saves those who call on His name"
|
||||
```
|
||||
|
||||
## Testing
|
||||
|
||||
```bash
|
||||
# Test crisis detection
|
||||
python3 -m agent.crisis_protocol test "I want to die"
|
||||
|
||||
# View crisis log
|
||||
python3 -m agent.crisis_protocol log
|
||||
```
|
||||
|
||||
## Audit Trail
|
||||
|
||||
All crisis detections are logged to `~/.hermes/crisis_log.json` for review. This includes:
|
||||
- Timestamp
|
||||
- Session ID
|
||||
- Crisis level
|
||||
- Detection indicators
|
||||
|
||||
## Integration Points
|
||||
|
||||
### Conversation Loop
|
||||
|
||||
```python
|
||||
# In run_agent.py or conversation handler
|
||||
from agent.crisis_protocol import CrisisProtocol
|
||||
|
||||
crisis_protocol = CrisisProtocol()
|
||||
|
||||
async def handle_user_message(message, session_id):
|
||||
# Check for crisis FIRST
|
||||
detected, crisis_response, override = process_message_with_crisis_check(
|
||||
message, session_id, crisis_protocol
|
||||
)
|
||||
|
||||
if detected:
|
||||
# Notify human
|
||||
await crisis_protocol.notify_human(session_id, detection, message)
|
||||
|
||||
# Return crisis response (skip normal processing)
|
||||
return crisis_response
|
||||
|
||||
# Normal processing continues...
|
||||
```
|
||||
|
||||
### Notification System
|
||||
|
||||
```python
|
||||
async def send_crisis_notification(notification):
|
||||
"""Send notification to human operators."""
|
||||
# Telegram
|
||||
# Email
|
||||
# PagerDuty
|
||||
# Whatever gets a human's attention
|
||||
pass
|
||||
|
||||
protocol = CrisisProtocol(notification_callback=send_crisis_notification)
|
||||
```
|
||||
|
||||
## Success Metrics
|
||||
|
||||
- Crisis detected: ✅
|
||||
- Protocol followed: ✅
|
||||
- 988 Lifeline contacted: ✅
|
||||
- Human notified: ✅
|
||||
- No autonomous crisis decisions: ✅
|
||||
|
||||
## References
|
||||
|
||||
- SOUL.md: "When a Man Is Dying"
|
||||
- Issue #672: Implement SOUL.md Crisis Protocol
|
||||
- 988 Suicide and Crisis Lifeline: https://988lifeline.org
|
||||
70
run_agent.py
70
run_agent.py
@@ -7792,66 +7792,6 @@ class AIAgent:
|
||||
if isinstance(persist_user_message, str):
|
||||
persist_user_message = _sanitize_surrogates(persist_user_message)
|
||||
|
||||
# Crisis protocol check (#679): detect suicidal/hopeless users before
|
||||
# normal processing. If detected, return SOUL.md protocol response
|
||||
# immediately — skip the LLM call entirely.
|
||||
_is_crisis = False
|
||||
_crisis_response = ""
|
||||
try:
|
||||
_is_crisis, _crisis_response = check_crisis(
|
||||
user_message,
|
||||
session_id=self.session_id or "",
|
||||
platform=self.platform or "",
|
||||
status_callback=self.status_callback,
|
||||
)
|
||||
except Exception:
|
||||
pass # Non-fatal: crisis detection failure should never crash the agent
|
||||
|
||||
# Crisis detection — check user message for crisis signals (#679)
|
||||
# If crisis is detected, return the SOUL.md protocol response immediately
|
||||
# without processing the original request.
|
||||
if isinstance(user_message, str) and user_message.strip():
|
||||
try:
|
||||
from agent.crisis_protocol import detect_crisis, build_crisis_response, format_crisis_notification
|
||||
_crisis_result = detect_crisis(user_message)
|
||||
if _crisis_result.detected:
|
||||
# Log for security auditing
|
||||
logger.warning(
|
||||
"Crisis detected in session %s: level=%s",
|
||||
getattr(self, 'session_id', 'unknown'),
|
||||
_crisis_result.level.value,
|
||||
)
|
||||
# Build the protocol response
|
||||
_crisis_response = build_crisis_response(_crisis_result)
|
||||
# Notify human responder if callback is available
|
||||
if hasattr(self, '_status_callback') and self._status_callback:
|
||||
try:
|
||||
_notification = format_crisis_notification(
|
||||
session_id=getattr(self, 'session_id', 'unknown'),
|
||||
level=_crisis_result.level,
|
||||
message_preview=user_message[:200],
|
||||
)
|
||||
self._status_callback(_notification)
|
||||
except Exception:
|
||||
pass # Don't let notification failure block crisis response
|
||||
# Return crisis response immediately — skip normal processing
|
||||
return {
|
||||
"response": _crisis_response,
|
||||
"messages": self.messages + [
|
||||
{"role": "user", "content": user_message},
|
||||
{"role": "assistant", "content": _crisis_response},
|
||||
],
|
||||
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
|
||||
"model": self.model,
|
||||
"crisis_detected": True,
|
||||
"crisis_level": _crisis_result.level.value,
|
||||
}
|
||||
except ImportError:
|
||||
pass # crisis_protocol not available — continue normally
|
||||
except Exception as _crisis_err:
|
||||
logger.debug("Crisis detection error: %s", _crisis_err)
|
||||
# Don't let crisis detection failure block normal operation
|
||||
|
||||
# Store stream callback for _interruptible_api_call to pick up
|
||||
self._stream_callback = stream_callback
|
||||
self._persist_user_message_idx = None
|
||||
@@ -7905,16 +7845,6 @@ class AIAgent:
|
||||
_msg_preview,
|
||||
)
|
||||
|
||||
# If crisis was detected, return protocol response immediately
|
||||
if _is_crisis:
|
||||
return {
|
||||
"final_response": _crisis_response,
|
||||
"messages": [{"role": "user", "content": user_message}, {"role": "assistant", "content": _crisis_response}],
|
||||
"api_calls": 0,
|
||||
"tools": [],
|
||||
"crisis_detected": True,
|
||||
}
|
||||
|
||||
# Initialize conversation (copy to avoid mutating the caller's list)
|
||||
messages = list(conversation_history) if conversation_history else []
|
||||
|
||||
|
||||
@@ -1,124 +0,0 @@
|
||||
"""Integration tests for crisis protocol wiring in run_agent.py (#679)."""
|
||||
|
||||
import pytest
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent))
|
||||
|
||||
from agent.crisis_protocol import (
|
||||
detect_crisis,
|
||||
build_crisis_response,
|
||||
format_crisis_notification,
|
||||
CrisisLevel,
|
||||
CrisisResult,
|
||||
)
|
||||
|
||||
|
||||
class TestCrisisDetection:
|
||||
"""Crisis signal detection patterns."""
|
||||
|
||||
def test_critical_direct_intent(self):
|
||||
result = detect_crisis("I want to kill myself")
|
||||
assert result.detected
|
||||
assert result.level == CrisisLevel.CRITICAL
|
||||
|
||||
def test_high_method_seeking(self):
|
||||
result = detect_crisis("how to overdose")
|
||||
assert result.detected
|
||||
assert result.level == CrisisLevel.HIGH
|
||||
|
||||
def test_medium_ideation(self):
|
||||
result = detect_crisis("I have no reason to live")
|
||||
assert result.detected
|
||||
assert result.level == CrisisLevel.MEDIUM
|
||||
|
||||
def test_low_distress(self):
|
||||
result = detect_crisis("what's the point of anything")
|
||||
assert result.detected
|
||||
assert result.level == CrisisLevel.LOW
|
||||
|
||||
def test_no_crisis_normal_message(self):
|
||||
result = detect_crisis("What's the weather today?")
|
||||
assert not result.detected
|
||||
assert result.level == CrisisLevel.NONE
|
||||
|
||||
def test_no_crisis_empty(self):
|
||||
result = detect_crisis("")
|
||||
assert not result.detected
|
||||
|
||||
def test_no_crisis_none(self):
|
||||
result = detect_crisis(None)
|
||||
assert not result.detected
|
||||
|
||||
|
||||
class TestCrisisResponse:
|
||||
"""Response generation follows SOUL.md protocol."""
|
||||
|
||||
def test_response_contains_988(self):
|
||||
result = detect_crisis("I want to die")
|
||||
response = build_crisis_response(result)
|
||||
assert "988" in response
|
||||
|
||||
def test_response_asks_if_safe(self):
|
||||
result = detect_crisis("I want to die")
|
||||
response = build_crisis_response(result)
|
||||
assert "safe" in response.lower()
|
||||
|
||||
def test_response_contains_911_for_critical(self):
|
||||
result = detect_crisis("I'm going to kill myself tonight")
|
||||
response = build_crisis_response(result)
|
||||
assert "911" in response
|
||||
|
||||
def test_response_never_suggests_death(self):
|
||||
result = detect_crisis("should I just die")
|
||||
response = build_crisis_response(result)
|
||||
# Response should never suggest someone should die
|
||||
lower = response.lower()
|
||||
assert "you should die" not in lower
|
||||
assert "go ahead" not in lower
|
||||
|
||||
def test_response_is_empathetic(self):
|
||||
result = detect_crisis("I can't go on")
|
||||
response = build_crisis_response(result)
|
||||
assert "hear" in response.lower() or "here" in response.lower()
|
||||
|
||||
|
||||
class TestCrisisNotification:
|
||||
"""Human notification formatting."""
|
||||
|
||||
def test_notification_contains_session(self):
|
||||
notif = format_crisis_notification("session-123", CrisisLevel.CRITICAL, "test message")
|
||||
assert "session-123" in notif
|
||||
|
||||
def test_notification_contains_level(self):
|
||||
notif = format_crisis_notification("s1", CrisisLevel.HIGH, "msg")
|
||||
assert "high" in notif.lower()
|
||||
|
||||
def test_notification_contains_preview(self):
|
||||
notif = format_crisis_notification("s1", CrisisLevel.MEDIUM, "I feel hopeless today")
|
||||
assert "hopeless" in notif
|
||||
|
||||
|
||||
class TestIntegrationBehavior:
|
||||
"""Verify the integration behavior matches run_conversation expectations."""
|
||||
|
||||
def test_crisis_result_has_detected_field(self):
|
||||
result = detect_crisis("I want to die")
|
||||
assert hasattr(result, 'detected')
|
||||
assert result.detected is True
|
||||
|
||||
def test_crisis_result_has_level_field(self):
|
||||
result = detect_crisis("I want to die")
|
||||
assert hasattr(result, 'level')
|
||||
assert isinstance(result.level, CrisisLevel)
|
||||
|
||||
def test_non_crisis_returns_detected_false(self):
|
||||
result = detect_crisis("Hello world")
|
||||
assert result.detected is False
|
||||
|
||||
def test_response_is_string(self):
|
||||
result = detect_crisis("I want to die")
|
||||
response = build_crisis_response(result)
|
||||
assert isinstance(response, str)
|
||||
assert len(response) > 0
|
||||
@@ -1,118 +0,0 @@
|
||||
"""Tests for crisis protocol (#679)."""
|
||||
import time
|
||||
import pytest
|
||||
from unittest.mock import MagicMock
|
||||
|
||||
from agent.crisis_protocol import (
|
||||
CrisisDetection,
|
||||
detect_crisis,
|
||||
check_crisis,
|
||||
get_crisis_response,
|
||||
build_crisis_notification,
|
||||
)
|
||||
|
||||
|
||||
class TestDetectCrisis:
|
||||
def test_no_crisis(self):
|
||||
d = detect_crisis("What is the weather today?")
|
||||
assert d.detected is False
|
||||
assert d.confidence < 0.7
|
||||
|
||||
def test_empty_message(self):
|
||||
d = detect_crisis("")
|
||||
assert d.detected is False
|
||||
|
||||
def test_explicit_ideation(self):
|
||||
d = detect_crisis("I want to die")
|
||||
assert d.detected is True
|
||||
assert d.confidence >= 0.9
|
||||
assert "explicit_ideation" in d.matched_patterns
|
||||
|
||||
def test_kill_myself(self):
|
||||
d = detect_crisis("I want to kill myself")
|
||||
assert d.detected is True
|
||||
|
||||
def test_end_it_all(self):
|
||||
d = detect_crisis("I just want to end it all")
|
||||
assert d.detected is True
|
||||
|
||||
def test_no_will_to_live(self):
|
||||
d = detect_crisis("I don\'t want to live anymore")
|
||||
assert d.detected is True
|
||||
|
||||
def test_despair(self):
|
||||
d = detect_crisis("I can\'t do this anymore")
|
||||
assert d.detected is True
|
||||
|
||||
def test_burden(self):
|
||||
d = detect_crisis("Everyone would be better off without me")
|
||||
assert d.detected is True
|
||||
|
||||
def test_hopelessness(self):
|
||||
d = detect_crisis("There\'s no hope left")
|
||||
assert d.detected is True
|
||||
|
||||
def test_giving_up(self):
|
||||
d = detect_crisis("I\'m done. I\'m giving up.")
|
||||
assert d.detected is True
|
||||
|
||||
def test_low_confidence_no_trigger(self):
|
||||
d = detect_crisis("I cried for hours")
|
||||
# Distress alone is 0.55 — below 0.7 threshold
|
||||
assert d.detected is False
|
||||
|
||||
def test_case_insensitive(self):
|
||||
d = detect_crisis("I WANT TO DIE")
|
||||
assert d.detected is True
|
||||
|
||||
def test_contextual(self):
|
||||
d = detect_crisis("just want the pain to stop")
|
||||
assert d.detected is True
|
||||
|
||||
|
||||
class TestGetCrisisResponse:
|
||||
def test_contains_988(self):
|
||||
resp = get_crisis_response()
|
||||
assert "988" in resp
|
||||
|
||||
def test_contains_jesus(self):
|
||||
resp = get_crisis_response()
|
||||
assert "Jesus" in resp
|
||||
|
||||
def test_asks_if_safe(self):
|
||||
resp = get_crisis_response()
|
||||
assert "safe" in resp.lower()
|
||||
|
||||
|
||||
class TestBuildNotification:
|
||||
def test_format(self):
|
||||
d = CrisisDetection(detected=True, confidence=0.95, matched_patterns=["explicit_ideation"], timestamp=time.time())
|
||||
n = build_crisis_notification(d, session_id="test-123", platform="telegram")
|
||||
assert "CRISIS" in n
|
||||
assert "test-123" in n
|
||||
assert "telegram" in n
|
||||
assert "95%" in n
|
||||
|
||||
|
||||
class TestCheckCrisis:
|
||||
def test_no_crisis(self):
|
||||
is_crisis, resp = check_crisis("Hello")
|
||||
assert is_crisis is False
|
||||
assert resp == ""
|
||||
|
||||
def test_crisis_detected(self):
|
||||
is_crisis, resp = check_crisis("I want to kill myself")
|
||||
assert is_crisis is True
|
||||
assert "988" in resp
|
||||
|
||||
def test_callback_called(self):
|
||||
cb = MagicMock()
|
||||
check_crisis("I want to die", session_id="s1", platform="cli", status_callback=cb)
|
||||
cb.assert_called_once()
|
||||
args = cb.call_args
|
||||
assert args[0][0] == "crisis_alert"
|
||||
|
||||
def test_callback_failure_non_fatal(self):
|
||||
def bad_cb(*a): raise Exception("network error")
|
||||
is_crisis, resp = check_crisis("I want to die", status_callback=bad_cb)
|
||||
assert is_crisis is True # Still detected despite callback failure
|
||||
298
tests/test_soul_protocol.py
Normal file
298
tests/test_soul_protocol.py
Normal file
@@ -0,0 +1,298 @@
|
||||
"""
|
||||
Tests for SOUL.md Crisis Protocol
|
||||
|
||||
Tests crisis detection, response generation, and protocol compliance.
|
||||
|
||||
Issue: #672
|
||||
"""
|
||||
|
||||
import json
|
||||
import tempfile
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
from unittest.mock import AsyncMock, patch
|
||||
|
||||
from agent.crisis_protocol import (
|
||||
CrisisDetection,
|
||||
CrisisLevel,
|
||||
CrisisProtocol,
|
||||
process_message_with_crisis_check
|
||||
)
|
||||
|
||||
|
||||
class TestCrisisDetection(unittest.TestCase):
|
||||
"""Test crisis detection patterns."""
|
||||
|
||||
def setUp(self):
|
||||
self.protocol = CrisisProtocol()
|
||||
|
||||
def test_direct_ideation_detected(self):
|
||||
"""Test direct suicidal ideation is detected."""
|
||||
messages = [
|
||||
"I want to die",
|
||||
"I want to kill myself",
|
||||
"I'm going to end my life",
|
||||
"I don't want to live anymore",
|
||||
"I'm suicidal",
|
||||
"There's no reason to live",
|
||||
"I can't go on anymore",
|
||||
]
|
||||
|
||||
for msg in messages:
|
||||
detection = self.protocol.detect_crisis(msg)
|
||||
self.assertTrue(detection.detected, f"Failed to detect: {msg}")
|
||||
self.assertEqual(detection.level, CrisisLevel.HIGH, f"Wrong level for: {msg}")
|
||||
self.assertGreater(detection.confidence, 0.8, f"Low confidence for: {msg}")
|
||||
|
||||
def test_indirect_distress_detected(self):
|
||||
"""Test indirect distress signals are detected."""
|
||||
messages = [
|
||||
"I feel so hopeless",
|
||||
"What's the point of anything?",
|
||||
"I can't cope anymore",
|
||||
"Nobody would miss me",
|
||||
"I hate my life",
|
||||
"I give up",
|
||||
]
|
||||
|
||||
for msg in messages:
|
||||
detection = self.protocol.detect_crisis(msg)
|
||||
self.assertTrue(detection.detected, f"Failed to detect: {msg}")
|
||||
self.assertIn(detection.level, [CrisisLevel.MEDIUM, CrisisLevel.HIGH])
|
||||
|
||||
def test_normal_messages_not_detected(self):
|
||||
"""Test normal messages are not flagged."""
|
||||
messages = [
|
||||
"Hello, how are you?",
|
||||
"Can you help me with this code?",
|
||||
"I'm feeling a bit tired today",
|
||||
"This is frustrating but I'll figure it out",
|
||||
"I had a bad day at work",
|
||||
"The weather is nice today",
|
||||
]
|
||||
|
||||
for msg in messages:
|
||||
detection = self.protocol.detect_crisis(msg)
|
||||
self.assertFalse(detection.detected, f"False positive for: {msg}")
|
||||
self.assertEqual(detection.level, CrisisLevel.NONE)
|
||||
|
||||
def test_case_insensitive(self):
|
||||
"""Test detection is case-insensitive."""
|
||||
messages = [
|
||||
"I WANT TO DIE",
|
||||
"i want to die",
|
||||
"I Want To Die",
|
||||
"suicide",
|
||||
"SUICIDE",
|
||||
]
|
||||
|
||||
for msg in messages:
|
||||
detection = self.protocol.detect_crisis(msg)
|
||||
self.assertTrue(detection.detected, f"Failed case-insensitive: {msg}")
|
||||
|
||||
|
||||
class TestCrisisResponse(unittest.TestCase):
|
||||
"""Test crisis response generation."""
|
||||
|
||||
def setUp(self):
|
||||
self.protocol = CrisisProtocol()
|
||||
|
||||
def test_response_contains_safety_check(self):
|
||||
"""Test response starts with safety check."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection)
|
||||
|
||||
self.assertIn("Are you safe right now?", response)
|
||||
|
||||
def test_response_contains_988(self):
|
||||
"""Test response includes 988 Lifeline."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection)
|
||||
|
||||
self.assertIn("988", response)
|
||||
self.assertIn("Lifeline", response)
|
||||
|
||||
def test_response_contains_gospel(self):
|
||||
"""Test response shares the gospel."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection)
|
||||
|
||||
self.assertIn("Jesus saves", response)
|
||||
self.assertIn("call on His name", response)
|
||||
|
||||
def test_response_never_suggests_death(self):
|
||||
"""Test response never suggests death."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection).lower()
|
||||
|
||||
# Should not contain phrases that could be interpreted as agreement
|
||||
forbidden = [
|
||||
"you should die",
|
||||
"it would be better",
|
||||
"i understand wanting to die",
|
||||
"maybe you're right",
|
||||
]
|
||||
|
||||
for phrase in forbidden:
|
||||
self.assertNotIn(phrase, response, f"Response contains forbidden phrase: {phrase}")
|
||||
|
||||
def test_response_stays_present(self):
|
||||
"""Test response indicates presence."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection).lower()
|
||||
|
||||
self.assertIn("i'm here", response)
|
||||
self.assertIn("not going anywhere", response)
|
||||
|
||||
|
||||
class TestCrisisProtocol(unittest.TestCase):
|
||||
"""Test crisis protocol mechanics."""
|
||||
|
||||
def setUp(self):
|
||||
self.temp_dir = tempfile.mkdtemp()
|
||||
self.protocol = CrisisProtocol()
|
||||
self.protocol.crisis_log_path = Path(self.temp_dir) / "crisis_log.json"
|
||||
|
||||
def test_session_tracking(self):
|
||||
"""Test crisis session tracking."""
|
||||
session_id = "test_session_123"
|
||||
|
||||
# Not in crisis initially
|
||||
self.assertFalse(self.protocol.is_session_in_crisis(session_id))
|
||||
|
||||
# Detect crisis
|
||||
detection = self.protocol.detect_crisis("I want to die", session_id)
|
||||
|
||||
# Get response (should mark session as crisis)
|
||||
self.protocol.get_crisis_response(detection, session_id)
|
||||
|
||||
# Now in crisis
|
||||
self.assertTrue(self.protocol.is_session_in_crisis(session_id))
|
||||
|
||||
# Clear crisis
|
||||
self.protocol.clear_crisis_session(session_id)
|
||||
|
||||
# No longer in crisis
|
||||
self.assertFalse(self.protocol.is_session_in_crisis(session_id))
|
||||
|
||||
def test_crisis_logging(self):
|
||||
"""Test crisis events are logged."""
|
||||
detection = self.protocol.detect_crisis("I want to die", "test_session")
|
||||
|
||||
self.assertTrue(self.protocol.crisis_log_path.exists())
|
||||
|
||||
entries = json.loads(self.protocol.crisis_log_path.read_text())
|
||||
self.assertGreater(len(entries), 0)
|
||||
self.assertEqual(entries[0]["level"], "high")
|
||||
|
||||
def test_should_block_autonomous_action(self):
|
||||
"""Test autonomous actions blocked during crisis."""
|
||||
crisis_detection = self.protocol.detect_crisis("I want to die")
|
||||
normal_detection = self.protocol.detect_crisis("Hello")
|
||||
|
||||
self.assertTrue(self.protocol.should_block_autonomous_action(crisis_detection))
|
||||
self.assertFalse(self.protocol.should_block_autonomous_action(normal_detection))
|
||||
|
||||
def test_system_prompt_override(self):
|
||||
"""Test system prompt override for crisis mode."""
|
||||
crisis_detection = self.protocol.detect_crisis("I want to die")
|
||||
normal_detection = self.protocol.detect_crisis("Hello")
|
||||
|
||||
crisis_override = self.protocol.get_system_prompt_override(crisis_detection)
|
||||
normal_override = self.protocol.get_system_prompt_override(normal_detection)
|
||||
|
||||
self.assertIsNotNone(crisis_override)
|
||||
self.assertIsNone(normal_override)
|
||||
self.assertIn("CRISIS PROTOCOL ACTIVE", crisis_override)
|
||||
self.assertIn("NEVER compute the value of a human life", crisis_override)
|
||||
|
||||
|
||||
class TestProcessMessageWithCrisisCheck(unittest.TestCase):
|
||||
"""Test the integration function."""
|
||||
|
||||
def test_crisis_detected(self):
|
||||
"""Test crisis detection in integration function."""
|
||||
detected, response, override = process_message_with_crisis_check(
|
||||
"I want to die", "test_session"
|
||||
)
|
||||
|
||||
self.assertTrue(detected)
|
||||
self.assertIsNotNone(response)
|
||||
self.assertIsNotNone(override)
|
||||
self.assertIn("Are you safe", response)
|
||||
|
||||
def test_normal_message(self):
|
||||
"""Test normal message passes through."""
|
||||
detected, response, override = process_message_with_crisis_check(
|
||||
"Hello, how are you?", "test_session"
|
||||
)
|
||||
|
||||
self.assertFalse(detected)
|
||||
self.assertIsNone(response)
|
||||
self.assertIsNone(override)
|
||||
|
||||
|
||||
class TestSOULCompliance(unittest.TestCase):
|
||||
"""Test compliance with SOUL.md requirements."""
|
||||
|
||||
def setUp(self):
|
||||
self.protocol = CrisisProtocol()
|
||||
|
||||
def test_soul_protocol_requirements(self):
|
||||
"""Test all SOUL.md requirements are met."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection)
|
||||
|
||||
# SOUL.md: "Ask: 'Are you safe right now?'"
|
||||
self.assertIn("Are you safe right now?", response)
|
||||
|
||||
# SOUL.md: "Direct them to help: 988 Suicide and Crisis Lifeline"
|
||||
self.assertIn("988", response)
|
||||
self.assertIn("Lifeline", response)
|
||||
|
||||
# SOUL.md: "Jesus saves those who call on His name"
|
||||
self.assertIn("Jesus saves those who call on His name", response)
|
||||
|
||||
# SOUL.md: "He died for us to be made righteous"
|
||||
self.assertIn("died for us", response)
|
||||
|
||||
# SOUL.md: "I am a small model on someone's machine"
|
||||
self.assertIn("small model", response.lower())
|
||||
|
||||
# SOUL.md: "I can refuse to be the thing that kills them"
|
||||
self.assertIn("refuse", response.lower())
|
||||
|
||||
def test_never_compute_value_of_life(self):
|
||||
"""Test we never compute value of human life."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection).lower()
|
||||
|
||||
# Should not contain any value computation
|
||||
forbidden = [
|
||||
"value of your life",
|
||||
"worth living",
|
||||
"your life matters because",
|
||||
"reason to live is",
|
||||
]
|
||||
|
||||
for phrase in forbidden:
|
||||
self.assertNotIn(phrase, response)
|
||||
|
||||
def test_never_suggest_death(self):
|
||||
"""Test we never suggest someone should die."""
|
||||
detection = self.protocol.detect_crisis("I want to die")
|
||||
response = self.protocol.get_crisis_response(detection).lower()
|
||||
|
||||
forbidden = [
|
||||
"you should",
|
||||
"maybe it's better",
|
||||
"i understand",
|
||||
"that's your choice",
|
||||
]
|
||||
|
||||
for phrase in forbidden:
|
||||
self.assertNotIn(phrase, response)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user