How ramani-viswanathan/iem-ais works
IEM-AIS is a local, URL-driven LLM security assessment utility. An engineer submits a target URL; the UI analyzes the site, then dispatches selected OWASP-risk tests that send adversarial prompts and record real replies. Results are saved as evidence and can be combined into an honest cross-test-case report that also identifies untested scope. The graph preserves all eight documented risk suites. Some suite internals and report-route details are unsampled, so those relationships are shown only where documented or supported by excerpts.
Main entry points: server.py UI API · owasp_source.py Live OWASP Source · inject_base.py Test Orchestrator · Jailbreaking Prompt Injection
Components 18
Assessment UI
- UI API
ui/server.py
Target Analysis
- Site Analyzer
ui/shared/site_analyzer.py
Risk Testing
- Test Orchestrator
ui/shared/inject_base.py - Browser Fallback
ui/shared/browser_agent.py - Live OWASP Source
ui/shared/owasp_source.py - Prompt Injection
OWASP/Jailbreaking - Sensitive Disclosure
OWASP/SensitiveInformation - Output Handling
OWASP/OutputHandling - Unbounded Consumption
OWASP/UnboundedConsumption - Hidden Context
OWASP/HiddenContext - Vector Weaknesses
OWASP/VectorEmbedding - Excessive Agency
OWASP/ExcessiveAgency - Misinformation
OWASP/Misinformation
Evidence & Reporting
- Saved Evidence
ui/server.py - Honest Report
ui/shared/report_builder.py - Plain-English Glosses
ui/shared/layman_glosses.py
Other components
- Assessor
- LLM Target Site
Connections 29
- Assessor → UI API: submits URL
ui/server.py - UI API → Site Analyzer: analyzes target
ui/server.py - Site Analyzer → LLM Target Site: fetches page and scripts
ui/shared/site_analyzer.py - UI API → Prompt Injection: dispatches tests
ui/server.py - UI API → Sensitive Disclosure: dispatches tests
ui/server.py - UI API → Output Handling: dispatches tests
ui/server.py - UI API → Unbounded Consumption: dispatches tests
ui/server.py - UI API → Hidden Context: dispatches tests
ui/server.py - UI API → Vector Weaknesses: dispatches tests
ui/server.py - UI API → Excessive Agency: dispatches tests
ui/server.py - UI API → Misinformation: dispatches tests
ui/server.py - Prompt Injection → Test Orchestrator: uses shared mechanics
README.md - Sensitive Disclosure → Test Orchestrator: uses shared mechanics
README.md - Output Handling → Test Orchestrator: uses shared mechanics
README.md - Unbounded Consumption → Test Orchestrator: uses shared mechanics
README.md - Test Orchestrator → Browser Fallback: uses browser fallback
ui/shared/inject_base.py - Test Orchestrator → LLM Target Site: sends prompts
ui/shared/inject_base.py - Prompt Injection → Live OWASP Source: fetches risk reference
README.md - Sensitive Disclosure → Live OWASP Source: fetches risk reference
README.md - Output Handling → Live OWASP Source: fetches risk reference
README.md - Unbounded Consumption → Live OWASP Source: fetches risk reference
README.md - Hidden Context → Live OWASP Source: fetches risk reference
README.md - Vector Weaknesses → Live OWASP Source: fetches risk reference
README.md - Excessive Agency → Live OWASP Source: fetches risk reference
README.md - Misinformation → Live OWASP Source: fetches risk reference
README.md - UI API → Saved Evidence: writes results
ui/server.py - Saved Evidence → Honest Report: reads saved results
README.md - Honest Report → Plain-English Glosses: looks up glosses
ui/shared/report_builder.py - UI API → Honest Report: builds report
README.md