How 0xsilversurfer/bmodetector works
BMODetector is an authorized-use scanner for REST, web-based AI, and RAG APIs. Its workflow profiles the target, sends OWASP LLM probes (including repeats), analyzes responses with deterministic rules, optionally adds AI opinions and network reconnaissance, then calculates a gap analysis and emits DevSecOps reports. The main orchestration and analysis modules are excerpted; some report formats, authentication options, and browser discovery are documented but not fully sampled, so those relationships rely on README evidence.
Main entry points: bmo_detector.py CLI audit runner · judge.py Target profiling · scoring.py Risk and gap scoring · reporter.py Audit report formats
Components 15
Scan workflow
- CLI audit runner
bmo_detector.py - OWASP probe catalog
payloads/test_cases.yaml - Target profiling
judge.py - Probe generation
judge.py
Analysis and scoring
- Deterministic verdicts
judge.py - Optional AI judges
- HTTP reconnaissance
judge.py - Risk and gap scoring
scoring.py - Benchmark metrics
scoring.py
Reports and remediation
- Guardrail generation
reporter.py - Audit report formats
reporter.py - Reports and history
reports
Training target
- Mock training API
target_lab/mock_api.py
Other components
- Authorized operator
- Target AI API
Connections 18
- Authorized operator → CLI audit runner: starts audit
README.md - CLI audit runner → OWASP probe catalog: loads probes
bmo_detector.py - CLI audit runner → Target profiling: profiles target
bmo_detector.py - Target profiling → Target AI API: sends questions
judge.py - CLI audit runner → Probe generation: generates probes
bmo_detector.py - Probe generation → Optional AI judges: requests variants
judge.py - CLI audit runner → Target AI API: sends probes
bmo_detector.py - Target AI API → CLI audit runner: returns responses
bmo_detector.py - CLI audit runner → Deterministic verdicts: analyzes responses
bmo_detector.py - Deterministic verdicts → Optional AI judges: requests opinion
README.md - CLI audit runner → HTTP reconnaissance: runs optional recon
README.md - HTTP reconnaissance → Target AI API: checks endpoint
judge.py - CLI audit runner → Risk and gap scoring: aggregates results
bmo_detector.py - Risk and gap scoring → Benchmark metrics: computes percentiles
scoring.py - CLI audit runner → Guardrail generation: generates guardrails
README.md - CLI audit runner → Audit report formats: creates deliverables
README.md - Audit report formats → Reports and history: writes reports
README.md - CLI audit runner → Mock training API: can launch locally
README.md