Give agents a bounded execution context.

Integration model

1. Human-defined intent: establish the mission and acceptance conditions.
2. Boundary: identify the responsibility and permitted change surface.
3. Contracts: provide the interfaces and invariants the agent must preserve.
4. Execution: give the agent only the tools, files, environments, and credentials required for that task.
5. Verification: require evidence before the result is accepted or promoted.

Proposed agent context

mission
owning_boundary
allowed_files
allowed_tools
dependencies
contracts
constraints
forbidden_assumptions
acceptance
verification_requirements

This is a conceptual field set, not a stable agent protocol. A future implementation can serialize it using the formal BOUND artifacts rather than inventing a separate agent-specific authority model.

Tool interaction

Tools should be exposed as capabilities inside the bounded context. The ability to call a tool is not itself permission to change every resource reachable through that tool.

Human accountability

The agent may implement, inspect, test, or propose changes. Ownership of intent, trade-offs, boundary changes, and acceptance remains explicitly assigned to the responsible human or governing process.