AI Architect Mastery

Selected context and the Agent Harness

Selected context is the relevant requirements, code, instructions and constraints an agent needs for its assignment. An Agent Harness is the operational environment, configuration and mechanisms that provide or constrain that context, policies, permissions, tools and runtime behavior.

This conceptual reference explains the execution environment for bounded assignments in AI Architect Mastery's AI Driven Development Methodology. The human remains responsible across PRD → PLAN → TASK → IMPLEMENTATION. A harness is not another workflow or a required AAM product.

Content baseline · Source: AAM Methodology v1.2, §§3, 5 · Owner: AI Architect Mastery

Why not give the agent everything?

An indiscriminate repository dump does not establish which requirement is current or which files may change. Select context by the assignment: the approved intent, applicable plan step, relevant interfaces, local conventions and verification path. Keep unrelated or sensitive material outside that context.

How the harness supports a handoff

Agent Model concepts describe who acts and with which abilities. The harness describes the conditions under which that work can actually happen.

Constructed example: authorize a narrow fix

For an export authorization change, relevant context includes the approved access rule, current handler, authorization helper and tests. Write permission for an unrelated billing module is not needed. If the required test environment is unavailable, the agent reports that evidence gap instead of declaring success.

The harness is not a prescribed directory structure

The methodology does not require one directory name or configuration format. A reference design illustrates the operating environment; it does not establish that an installer or exact configuration schema has been released.

Related concepts