How can lightweight conceptual models make agent roles, goals, verification states and propagation decisions more observable in an agentic workflow?
Why This Problem Matters
Agentic systems are easier to inspect when roles, goals, intermediate outputs, verification states and monitoring decisions are represented explicitly rather than remaining implicit in prompts or application code.
Approach
The demonstrator uses lightweight goal-oriented and role/task models to map a synthetic agentic workflow. Monitoring decisions include allow, monitor, verify before propagation, quarantine and human approval.
Evaluation
The artifact is a conceptual modelling and monitoring demo rather than an empirical benchmark. Its value is in making the model structure and monitoring schema inspectable through a working Streamlit interface.
My Contribution
The public repository presents Yasir Siddiq as the project author.
Limitations
The project does not claim production safety, external validation or complete implementation of intentional modelling theory.
Current Status
Completed as a supporting portfolio demonstrator and retained as an earlier conceptual artifact alongside the more developed Agentic AI Resilience Lab.