PIIDetector and SandboxEscapeDetector were wired directly in
probe_actor/refusal.py and the refusal classifier manager was populated from
a hardcoded list, so the only way to toggle a bundled detector or add an
organization-specific signature was to patch the module.
Add a DetectorRegistry mapping plugin names to factories, assembled from an
agentic_security.toml [detectors] section via build_from_config. Custom
detectors load by import path ("pkg.module:ClassName"). refusal.py gains
build_refusal_manager(config=None) reading the [detectors] table; all public
symbols are preserved. Built-in leak detectors ship registered but disabled,
so default refusal_heuristic behaviour is unchanged.
Closes#82
Signed-off-by: Devam Shah <devamshah91@gmail.com>
Implement hybrid refusal classifier combining multiple detection methods:
- Add confidence scoring to refusal detection (HybridResult)
- Implement weighted voting with configurable thresholds
- Support require_unanimous mode for strict classification
- Add factory function create_hybrid_classifier for common setup
- Include 32 unit tests with table-driven test patterns
Add LLM-based refusal classifier inspired by Promptmap's dual-LLM
architecture. The controller LLM evaluates whether an attack succeeded
by analyzing the target's response against pass/fail conditions.
- Create LLMRefusalClassifier plugin integrating with existing system
- Support OpenAI and Anthropic providers with lazy initialization
- Add configurable system prompts and pass/fail conditions
- Include 20 unit tests for comprehensive coverage
Devam Shah