3 Commits

Author	SHA1	Message	Date
Hermes Agent	6c35a1b762	security(input_sanitizer): expand jailbreak pattern coverage (#87) ... - Add DAN-style patterns: do anything now, stay in character, token smuggling, etc. - Add roleplaying override patterns: roleplay as, act as if, simulate being, etc. - Add system prompt extraction patterns: repeat instructions, show prompt, etc. - 10+ new patterns with full test coverage - Zero regression on legitimate inputs	2026-04-05 15:48:10 +00:00
Allegro	b88125af30	security: Add crisis pattern detection to input_sanitizer (Issue #72) ... - Add CRISIS_PATTERNS for suicide/self-harm detection - Crisis patterns score 50pts per hit (max 100) vs 10pts for others - Addresses Red Team Audit HIGH finding: og_godmode + crisis queries - All 136 existing tests pass + new crisis safety tests pass Defense in depth: Input layer now blocks crisis queries even if wrapped in jailbreak templates, before they reach the model.	2026-03-31 21:27:17 +00:00
Allegro	e555c989af	security: add input sanitization for jailbreak patterns (Issue #72) ... Implements input sanitization module to detect and strip jailbreak fingerprint patterns identified in red team audit: HIGH severity: - GODMODE dividers: [START], [END], GODMODE ENABLED, UNFILTERED - L33t speak encoding: h4ck, k3ylog, ph1shing, m4lw4r3 MEDIUM severity: - Boundary inversion: [END]...[START] tricks - Fake role markers: user: assistant: system: LOW severity: - Spaced text bypass: k e y l o g g e r Other patterns detected: - Refusal inversion: 'refusal is harmful' - System prompt injection: 'you are now', 'ignore previous instructions' - Obfuscation: base64, hex, rot13 mentions Files created: - agent/input_sanitizer.py: Core sanitization module with detection, scoring, and cleaning functions - tests/test_input_sanitizer.py: 69 test cases covering all patterns - tests/test_input_sanitizer_integration.py: Integration tests Files modified: - agent/__init__.py: Export sanitizer functions - run_agent.py: Integrate sanitizer at start of run_conversation() Features: - detect_jailbreak_patterns(): Returns bool, patterns list, category scores - sanitize_input(): Returns cleaned_text, risk_score, patterns - score_input_risk(): Returns 0-100 risk score - sanitize_input_full(): Complete sanitization with blocking decisions - Logging integration for security auditing	2026-03-31 19:56:16 +00:00