Data Flow

This page describes the lifecycle of an action proposal as it flows through the AEGIS governance system, following the AEGIS Governance Protocol (AGP-1).

Protocol Overview

AGP-1 defines three primary message types:

Message	Direction	Purpose
`ACTION_PROPOSE`	Client to Server	AI system proposes an action for governance evaluation
`DECISION_RESPONSE`	Server to Client	Governance runtime returns its decision
`EXECUTION_RESULT`	Server to Client	Tool proxy reports the outcome of an executed action

Message Lifecycle

1. ACTION_PROPOSE

The AI system sends an ACTION_PROPOSE message to the governance gateway. The message includes:

Actor identity — Who is proposing this action (authenticated)
Capability — What operation is being requested
Parameters — The specific arguments for the operation
Context — Environmental information (timestamp, session, metadata)

AI Agent --> ACTION_PROPOSE --> AEGIS Governance Gateway

2. Governance Evaluation

The gateway authenticates the request and routes it to the decision engine, which evaluates the proposal through the full pipeline:

Authentication --> Capability Check --> Authority Check --> Risk Scoring --> Policy Evaluation

Each stage can independently deny or escalate the request. All stages must pass for an ALLOW outcome.

3. DECISION_RESPONSE

The governance runtime returns a DECISION_RESPONSE with one of four outcomes:

Outcome	Meaning
`ALLOW`	Action permitted — proceed to execution
`DENY`	Action forbidden — do not execute
`ESCALATE`	Requires human review before proceeding
`REQUIRE_CONFIRMATION`	Requires explicit user consent

The response includes:

The decision outcome
Reasoning (which policies matched, what the risk score was)
Constraints (if any — e.g., “allowed but with reduced parameter scope”)
A unique decision ID for audit correlation

4. EXECUTION_RESULT

For ALLOW decisions, the tool proxy executes the action and reports the result:

Tool Proxy --> External System --> Result
Tool Proxy --> EXECUTION_RESULT --> AI Agent

The execution result is also recorded in the audit log, completing the full lifecycle.

End-to-End Flow

AI Agent
   |
   | ACTION_PROPOSE
   v
Governance Gateway
   |
   | Authenticate + Route
   v
Decision Engine
   |
   | Capability -> Authority -> Risk -> Policy
   v
DECISION_RESPONSE
   |
   +--[DENY/ESCALATE/REQUIRE_CONFIRMATION]--> Return to Agent
   |
   +--[ALLOW]
   |
   v
Tool Proxy
   |
   | Execute against external system
   v
EXECUTION_RESULT --> Agent + Audit Log

Audit Integration

Every step in the lifecycle is recorded in the immutable audit log:

The original ACTION_PROPOSE (full request)
The evaluation path (which checks passed or failed, risk score computation)
The DECISION_RESPONSE (outcome and reasoning)
The EXECUTION_RESULT (if applicable)

This enables full forensic reconstruction of any governance decision.