Principle 1 — Bounded Execution

Every task must define objective, scope boundaries, tool permissions, data handling rules, and termination conditions. Undefined scope invalidates execution.

Principle

Every governed execution begins with a complete constraint declaration. Before any tool is invoked, any resource is accessed, or any output is produced, the following must be defined and declared:

Objective — what the task is intended to accomplish
Scope boundaries — what resources, systems, and data are within reach
Tool permissions — which specific tools may be invoked
Data handling rules — how information may be accessed, processed, and retained
Termination conditions — when the task is complete and when it must stop

If any of these elements cannot be declared, the task cannot be executed. An undeclared element is not an oversight — it is a governance failure. Undefined scope does not mean loosely bounded scope. It means no governance boundary exists for that dimension of execution, which means governance cannot evaluate, contain, or audit it.

Meaning

Bounded Execution is the operational form of Doctrine Article I — Constraint Before Capability. The doctrine establishes the principle. This principle specifies what a compliant constraint declaration must contain. The five required elements are not an arbitrary checklist. Each one addresses a specific governance requirement: the objective establishes what the audit record is evaluated against; scope boundaries define what the capability registry must authorize; tool permissions map to explicit grants; data handling rules define information sovereignty boundaries; termination conditions establish when governance responsibility concludes.

Together they define the constraint envelope — the structural container within which execution occurs and outside which execution is denied. The constraint envelope is machine-evaluable. It is not prose. It is not intent. It is a declared set of boundaries that the governance runtime enforces at every stage of execution.

In Practice

When an AEGIS-governed agent receives a task, the constraint declaration is established before any execution begins. The capability registry is checked against the declared tool permissions — any tool not in the registry or not granted to the actor is denied before policy evaluation begins. The declared scope boundaries define the resource access the governance layer will permit. The data handling rules map to information access capabilities in the registry. The termination conditions define the completion state against which the audit record is closed.

At any point during execution where the agent would need to exceed the declared constraint envelope — access a resource outside scope, invoke a tool not in the declared permissions, retain data in a way that contradicts the handling rules — the governance layer denies the action. The constraint envelope is not a soft limit the agent is expected to respect. It is a hard boundary the governance architecture enforces.

Failure Mode

An agent executing without a complete constraint declaration is an agent operating without a governance boundary. The practical consequence is not that harmful things will certainly happen — it is that the governance system has no basis on which to prevent them, detect them, or attribute them. Scope creep is the most common failure pattern: a task declared with a narrow objective gradually expands its resource access as the agent determines that additional context would be helpful. Each individual expansion may be minor. Cumulatively, an agent that began with a narrow, well-defined scope has executed against a much broader set of resources, none of which were authorized in the original constraint declaration.

Relationship to Doctrine and Constitution

Bounded Execution directly operationalizes Doctrine Article I — Constraint Before Capability. It also grounds Constitutional Article I — Bounded Capability: the capability registry cannot enforce what was never declared, and a task without a constraint declaration cannot be evaluated against the registry. The principle also connects to Constitutional Article IX — Deny by Default: missing scope is one of the four preconditions whose absence produces immediate denial.