How praval2006/agentguard works
AgentGuard is an acceptance-verification layer for AI-built changes. Its principal reviewed workflow proposes explicit and inferred behaviours, lets a human decide which inferred behaviours to accept, grounds the selected scenarios into supported checks, executes them, and deterministically reports evidence and verdicts. The Python backend is distinct from the React presentation: the web demo plays authored fixtures and does not invoke the backend. Edges reflect the README and source references shown; unsampled details are omitted where relationships cannot be established.
Main entry points: acceptance.py Execution dispatcher · reviewed_workflow.py Reviewed workflow · grounding.py Scenario grounding · App.tsx React app router
Components 25
Intent and review
- Acceptance planner
agentguard/planner.py - OpenAI provider
agentguard/providers/openai_provider.py - Reviewed workflow
agentguard/reviewed_workflow.py - Acceptance contract
agentguard/acceptance_contract.py - CLI workflow
agentguard/cli.py
Grounding and verification
- Scenario grounding
agentguard/grounding.py - Scenario validation
agentguard/scenarios.py - Input derivation
agentguard/derivations.py - Execution dispatcher
agentguard/acceptance.py - HTTP executor
agentguard/http_execution.py - HTTP sequences
agentguard/sequence_execution.py - Trusted test executor
agentguard/execution.py - Registered checks
agentguard/check_registry.py - Coverage authorization
agentguard/coverage_authorization.py - Registered executor
agentguard/registered_execution.py - Deterministic verifier
agentguard/verifier.py - Acceptance report
agentguard/acceptance_report.py
Web presentation
- React app router
frontend/src/App.tsx - Product presentation
frontend/src/PresentationPage.tsx - Controlled review demo
frontend/src/ReviewDemo.tsx - Authored demo fixtures
frontend/src/accountDemoData.ts - Evidence and correction views
frontend/src/EvidenceDetail.tsx
Other components
- Task author
- Human reviewer
- Web visitor
Connections 24
- Task author → CLI workflow: provides task
README.md - CLI workflow → Reviewed workflow: starts reviewed flow
README.md - Reviewed workflow → Acceptance planner: requests proposal
agentguard/reviewed_workflow.py - Acceptance planner → OpenAI provider: requests reasoning
README.md - Reviewed workflow → Acceptance contract: creates contract
agentguard/reviewed_workflow.py - Human reviewer → Reviewed workflow: submits decisions
README.md - Reviewed workflow → Scenario grounding: grounds selections
agentguard/reviewed_workflow.py - Scenario grounding → Scenario validation: validates scenarios
agentguard/grounding.py - Scenario grounding → Input derivation: uses derivations
agentguard/grounding.py - Reviewed workflow → Execution dispatcher: runs verification
agentguard/reviewed_workflow.py - Execution dispatcher → Scenario validation: validates inputs
agentguard/acceptance.py - Execution dispatcher → HTTP executor: dispatches HTTP
agentguard/acceptance.py - Execution dispatcher → HTTP sequences: dispatches sequences
agentguard/acceptance.py - Execution dispatcher → Trusted test executor: dispatches tests
agentguard/acceptance.py - Execution dispatcher → Coverage authorization: checks authorization
agentguard/acceptance.py - Coverage authorization → Registered checks: resolves checks
agentguard/coverage_authorization.py - Execution dispatcher → Registered executor: runs authorized check
agentguard/acceptance.py - Execution dispatcher → Deterministic verifier: evaluates observations
agentguard/acceptance.py - Execution dispatcher → Acceptance report: projects outcome
README.md - Web visitor → React app router: opens experience
frontend/src/App.tsx - React app router → Product presentation: routes presentation
frontend/src/App.tsx - React app router → Controlled review demo: routes demo
frontend/src/App.tsx - Controlled review demo → Authored demo fixtures: selects fixture
frontend/src/ReviewDemo.tsx - Controlled review demo → Evidence and correction views: renders evidence
frontend/src/ReviewDemo.tsx