How thedead91/groundrumble works
GroundRumble is a client-side LLM security workbench. Its main workflow turns research sources into editable, ATLAS-mapped tests, then runs the same suite against a model lineup and presents inspectable verdicts and comparisons. The graph also captures the documented matrix, prompt editing, dashboard, sandbox, and credential-vault capabilities. README claims guide several subsystem links; sampled source confirms the app shell, generation orchestration, provider request handling, and judge-merge flow, but not every view’s internal wiring.
Main entry points: App.jsx App shell · useAIGeneration.js Generation workflow · useAuditRun.js Audit orchestration · DashboardView.jsx Security Dashboard
Components 24
Workbench UI
- App shell
src/App.jsx - Auditor Runner
src/components/views/RunnerView.jsx - Test Library
src/components/views/TestsView.jsx - ATLAS Matrix
src/components/views/MatrixView.jsx - Security Dashboard
src/components/views/DashboardView.jsx
Research To Tests
- Generation workflow
src/hooks/useAIGeneration.js - AI operations
src/utils/api/index.js - ATLAS catalog
src/data/atlas-bundled.js
Audit And Results
- Audit orchestration
src/hooks/useAuditRun.js - Verdict judging
src/utils/ai-judge.js - Audit history
src/context/HistoryContext.jsx - Reports and metrics
src/utils/report-builder.js
Models And Prompts
- Provider settings
src/context/ProvidersContext.jsx - Provider client
src/utils/api/provider-client.js - Judge client
src/utils/api/judge-client.js - Editable prompts
src/components/views/PromptsView.jsx - Sandbox models
src/utils/api/index.js - App settings
src/context/SettingsContext.jsx
Browser Data
- Browser storage
src/utils/storage.js - Credential vault
src/utils/vault.js
Other components
- Researcher
- Research sources
- Model endpoints
- IndexedDB
Connections 27
- Researcher → App shell: uses
- App shell → Auditor Runner: shows
- App shell → Test Library: shows
- App shell → ATLAS Matrix: shows
- App shell → Security Dashboard: shows
- Researcher → Research sources: provides
- Test Library → Generation workflow: starts generation
- Generation workflow → Research sources: fetches excerpts
- Generation workflow → AI operations: requests analysis
- Generation workflow → ATLAS catalog: maps techniques
- Generation workflow → Test Library: offers tests
- ATLAS Matrix → ATLAS catalog: browses
- Auditor Runner → Audit orchestration: starts audit
- Audit orchestration → Provider client: sends test prompts
- Audit orchestration → Verdict judging: evaluates responses
- Verdict judging → Judge client: requests judge
- Provider client → Model endpoints: calls endpoint
- Judge client → Model endpoints: calls endpoint
- Audit orchestration → Audit history: records results
- Security Dashboard → Audit history: reads runs
- Security Dashboard → Reports and metrics: builds reports
- Editable prompts → App settings: edits overrides
- Provider settings → Browser storage: persists config
- App settings → Browser storage: persists settings
- Credential vault → IndexedDB: stores credentials
- Provider settings → Credential vault: uses credentials
- Audit orchestration → Sandbox models: uses in sandbox