pair-programming-with-agent

What problem does it solve?

Prevents “vibe coding” pair sessions where an AI agent over-engineers, takes the wrong path, or invents domain assumptions instead of executing within human-defined constraints.

Core Features & Use Cases

• Human–Agent Navigation Framework: Establishes a clear mental model (human as navigator, agent as pilot) to guide collaboration and when the human must take control.
• Akita Capability Guardrails: Uses a “does well vs does badly” table to decide when to delegate tasks and when to require active human judgment.
• Interruption Triggers with Corrective Phrases: Provides concrete stop conditions for over-engineering, off-plan work, insufficient context, and disguised architectural decisions.
• Domain Context Injection Templates: Shows how to supply non-obvious, project-specific information (rules, workarounds, infra limits) before the agent proceeds.

Quick Start

Tell your agent: “Start implementing this task, but stop immediately if you propose premature abstractions or change the plan, and ask me for domain context before continuing.”