PV-PP Static Governance Sidecar
A packet-driven proof-of-concept for applying static governance rules to proposed agent actions before tool execution.
Overview
This repository presents a public draft white paper for a PV-PP static governance sidecar. The core idea is that agent tool layers can identify what actions are technically available, while a governance sidecar evaluates whether a proposed action is admissible before it happens.
The proof-of-concept uses a common runtime with scenario-specific packets. A scenario packet defines
the proposed action structure, governance predicates, decision rules, scoring policy, and trace
requirements. The runtime evaluates proposed actions and returns a decision such as
PROCEED, BLOCK, ESCALATE, or DEFER.
Available vs. Viable
Tool registries expose what an agent can call. A sidecar gate asks whether a proposed action should proceed.
Common Runtime
The runtime is not email-specific. The scenario packet supplies the governance rules and action structure.
Traceable Decisions
Each decision returns predicate-level trace information so a blocked or allowed action can be reviewed.
Demonstration Scenario
The white paper uses a batch email attachment scenario. An agent proposes multiple outbound email actions. A scenario packet defines a confidentiality rule that blocks a proposed send when an attachment contains an Amundsen R&D confidentiality marker. Clean sends may proceed; violating sends are blocked with a trace.
This is not a product announcement and not a production security layer. It is a proof-of-concept showing how static governance rules can be attached to proposed agent actions through a sidecar-style gate.