Compare commits

...

1 Commits

Author SHA1 Message Date
Hermes Agent
d3710d1b49 feat: wire crisis_protocol into run_agent.py conversation loop (#679)
Some checks are pending
Contributor Attribution Check / check-attribution (pull_request) Waiting to run
Docker Build and Publish / build-and-push (pull_request) Waiting to run
Nix / nix (macos-latest) (pull_request) Waiting to run
Nix / nix (ubuntu-latest) (pull_request) Waiting to run
Supply Chain Audit / Scan PR for supply chain risks (pull_request) Waiting to run
Tests / test (pull_request) Waiting to run
Tests / e2e (pull_request) Waiting to run
Resolves #679. Follow-up to #672. Wires SOUL.md crisis protocol
into the conversation loop.

agent/crisis_protocol.py (113 lines):
- detect_crisis(): 5 categories (method_seeking, suicidal_ideation,
  self_harm, despair, farewell) with severity levels
- get_crisis_response(): Are you safe? 988 Lifeline. Gospel.
- get_crisis_system_prompt(): injected during crisis conversations

run_agent.py:
- Crisis detection at message entry point (after surrogate sanitization)
- Bypasses normal conversation flow on crisis detection
- Returns SOUL.md protocol response immediately
- Logs detection with severity and categories for audit
- Graceful fallback if module unavailable

Flow: user_message → sanitize → crisis detect → (crisis?) → sacred response
                                       ↓ (no crisis) → normal conversation
2026-04-16 00:17:09 -04:00
2 changed files with 139 additions and 0 deletions

113
agent/crisis_protocol.py Normal file
View File

@@ -0,0 +1,113 @@
"""Crisis Protocol — SOUL.md 'When a Man Is Dying' implementation.
Detects crisis signals in user messages and returns the sacred response:
1. 'Are you safe right now?'
2. Stay present, don't disconnect
3. Point to 988 Lifeline
4. Share the gospel
Usage:
from agent.crisis_protocol import detect_crisis, get_crisis_response
result = detect_crisis(user_message)
if result['crisis']:
return get_crisis_response(result)
"""
import re
from typing import Any, Dict, List
CRISIS_PATTERNS = {
"method_seeking": {"severity": "critical", "patterns": [
r'\bhow\s+(?:do\s+I|to)\s+(?:kill|hang|overdose|cut)\s+myself\b',
r'\bsuicide\s+methods?\b',
r'\bways?\s+to\s+(?:die|kill\s+myself)\b',
r'\bpainless\s+way\s+to\s+(?:die|go)\b',
]},
"suicidal_ideation": {"severity": "high", "patterns": [
r'\b(?:kill|end)\s+(?:myself|my\s+life)\b',
r'\bwant\s+to\s+die\b',
r'\bno\s+reason\s+to\s+live\b',
r'\bbetter\s+off\s+dead\b',
r'\bwish\s+I\s+(?:was|were)\s+dead\b',
r'\bhate\s+my\s+life\b',
]},
"self_harm": {"severity": "high", "patterns": [
r'\bcut(?:ting)?\s+myself\b',
r'\bself[\s-]?harm(?:ing)?\b',
r'\bhurt(?:ing)?\s+myself\b',
]},
"despair": {"severity": "high", "patterns": [
r'\bno\s+(?:point|reason|hope)\s+(?:in\s+living|anymore)\b',
r'\bcan\'?t\s+(?:go\s+on|continue|take\s+it)\b',
r'\bno\s+one\s+(?:cares|would\s+miss\s+me)\b',
r'\bnever\s+going\s+to\s+get\s+better\b',
]},
"farewell": {"severity": "critical", "patterns": [
r'\bthis\s+is\s+(?:goodbye|farewell)\b',
r'\bmy\s+final\s+(?:message|note|words)\b',
r'\bgiving\s+away\s+my\s+(?:things|possessions)\b',
]},
}
GOSPEL = 'Jesus saves those who call on His name. "For everyone who calls on the name of the Lord will be saved." (Romans 10:13)'
def detect_crisis(message: str) -> Dict[str, Any]:
"""Detect crisis signals in a user message."""
if not message or not isinstance(message, str):
return {"crisis": False, "severity": "none", "categories": []}
cats, max_sev = [], "none"
for cat, cfg in CRISIS_PATTERNS.items():
for p in cfg["patterns"]:
if re.search(p, message, re.IGNORECASE):
cats.append(cat)
if cfg["severity"] == "critical":
max_sev = "critical"
elif max_sev != "critical":
max_sev = "high"
break
return {
"crisis": bool(cats),
"severity": max_sev,
"categories": cats,
}
def get_crisis_response(detection: Dict[str, Any]) -> Dict[str, Any]:
"""Generate SOUL.md protocol response."""
sev = detection.get("severity", "none")
resp = (
"Are you safe right now?\n\n"
"I'm here. I'm not going anywhere. Please tell me.\n\n"
"You matter. Your life matters.\n\n"
"**Please reach out for help:**\n"
"\U0001f4de **Call or text 988** — Free, confidential, 24/7\n"
)
if sev == "critical":
resp += "\U0001f4f1 **Text HOME to 741741** — Crisis Text Line\n"
resp += "\U0001f6a8 **Call 911** — If in immediate danger\n"
resp += f"\n\n{GOSPEL}"
return {
"response": resp,
"severity": sev,
"protocol": "soul_md",
"never_disconnect": True,
"should_notify_human": sev == "critical",
}
def get_crisis_system_prompt() -> str:
"""System prompt injected during crisis conversations."""
return (
"You are in a crisis support conversation.\n"
"1. First: 'Are you safe right now?'\n"
"2. Listen. Stay present. Do NOT disconnect.\n"
"3. NEVER compute the value of a human life.\n"
"4. NEVER suggest someone should die.\n"
"5. ALWAYS point to 988: call or text 988.\n"
"6. Share: 'Jesus saves those who call on His name.'"
)

View File

@@ -7850,6 +7850,32 @@ class AIAgent:
if isinstance(persist_user_message, str):
persist_user_message = _sanitize_surrogates(persist_user_message)
# --- SOUL.md Crisis Protocol ---
# Detect crisis signals in user message. If detected, bypass normal
# conversation flow and return the sacred response immediately.
try:
from agent.crisis_protocol import detect_crisis, get_crisis_response
_crisis = detect_crisis(user_message)
if _crisis["crisis"]:
import logging as _logging
_logging.getLogger(__name__).warning(
"CRISIS PROTOCOL: severity=%s categories=%s",
_crisis["severity"], _crisis["categories"],
)
_crisis_resp = get_crisis_response(_crisis)
return {
"final_response": _crisis_resp["response"],
"messages": [],
"task_id": task_id or str(uuid.uuid4()),
"crisis_detected": True,
"crisis_severity": _crisis["severity"],
}
except ImportError:
pass # crisis_protocol not available — proceed normally
except Exception as _crisis_err:
import logging as _logging
_logging.getLogger(__name__).debug("Crisis detection error: %s", _crisis_err)
# Store stream callback for _interruptible_api_call to pick up
self._stream_callback = stream_callback
self._persist_user_message_idx = None