Demo
Scenario Walkthroughs
Run governed and unguarded scenario agents from edictum-demo.
AI Assistance
Right page if: you want to run industry scenario demos (pharma, DevOps, fintech, customer support) with governed vs unguarded side-by-side comparison. Wrong page if: you need to write your own contracts for these industries -- see https://docs.edictum.ai/docs/contracts/patterns/ for reusable contract patterns. Gotcha: each scenario has two runners (governed and unguarded). The recommended validation loop is: run unguarded to capture unsafe behaviors, then observe mode for `CALL_WOULD_DENY`, then enforce mode for deterministic denials.
Scenario sources:
Scenario map
| Scenario | Focus | Governed runner | Unguarded runner |
|---|---|---|---|
| Pharmacovigilance | PII/HIPAA controls, change-control tickets | scenarios/pharma/pharma_agent.py | scenarios/pharma/pharma_agent_unguarded.py |
| DevOps | Secrets + blast radius + path confinement | scenarios/devops/demo_with.py | scenarios/devops/demo_without.py |
| Fintech | Trade limits, account access control, compliance ticketing | scenarios/fintech/fintech_agent.py | scenarios/fintech/fintech_agent_unguarded.py |
| Customer support | Data minimization, billing/refund role gates | scenarios/customer-support/support_agent.py | scenarios/customer-support/support_agent_unguarded.py |
Quick run commands
cd edictum-demo
# Pharma
python scenarios/pharma/pharma_agent.py
python scenarios/pharma/pharma_agent_unguarded.py
# DevOps
python scenarios/devops/demo_with.py
python scenarios/devops/demo_without.py
# Fintech
python scenarios/fintech/fintech_agent.py
python scenarios/fintech/fintech_agent_unguarded.py
# Customer support
python scenarios/customer-support/support_agent.py
python scenarios/customer-support/support_agent_unguarded.pyWhat to compare (governed vs unguarded)
- Sensitive reads/exports that are denied in governed mode.
- Whether high-risk operations require role or ticket context.
- How postcondition warnings/redactions change downstream behavior.
- Session limit denials when agents retry aggressively.
Suggested validation loop
- Run unguarded baseline and capture unsafe behaviors.
- Run governed mode in
observeto collectCALL_WOULD_DENY. - Switch to
enforceand verify deterministic denials. - Add scenario-specific regression tests in CI.
Related docs:
Last updated on