RI3 — Executable Governance
Governance Proxy
Paste a guardrails JSON config. Submit an agent action. See the decision: allow, deny, or gate. With a full trace of which persona and rule triggered it.
⚙
Guardrails Config
Gatekeep (ticketyboo.dev)
Simple Security
Custom JSON
→
Agent Action
▶
Evaluation Result
Submit an action to see the governance decision.
What this demonstrates
Governance as code
Rules are data. The guardrails config is JSON. The evaluation engine is 200 lines of Python. Changing governance policy is a config change, not a code change.
Rules are data. The guardrails config is JSON. The evaluation engine is 200 lines of Python. Changing governance policy is a config change, not a code change.
Deterministic evaluation
No LLM involved. Every evaluation of the same action against the same config produces the same result. Governance is predictable and auditable.
No LLM involved. Every evaluation of the same action against the same config produces the same result. Governance is predictable and auditable.
Persona dispatch
Rules are grouped by persona (Security Sentinel, Cost Guardian). Each persona has a different concern. The evaluation trace shows which persona's rule fired.
Rules are grouped by persona (Security Sentinel, Cost Guardian). Each persona has a different concern. The evaluation trace shows which persona's rule fired.
Three decisions
Allow: proceed. Deny: block with reason. Gate: pause for human approval. Gate is the most useful decision for production deployments and destructive operations.
Allow: proceed. Deny: block with reason. Gate: pause for human approval. Gate is the most useful decision for production deployments and destructive operations.
Read the companion article: Building a governance proxy in 200 lines of Python