What problem does it solve? Non-trivial code changes often fail because AI agents jump straight into editing without understanding blast radius, existing patterns, or scope boundaries. This Skill enforces a strict separation between read-only planning and approved execution so changes stay predictable and reviewable. ## Core Features & Use Cases - Intent Gating: Explicitly separates PLAN mode (research, no file changes) from EXECUTE mode (implement only the approved plan), preventing scope creep. - Impact-Aware Planning: Uses GitNexus queries (context, query, impact, detect_changes) to map blast radius and compare at least two implementation approaches before committing. - Verification & Reporting: Runs real test verification, linting, scope checks via detect_changes, and produces a deviation report after execution. - Use Case: When asked to add rate limiting to an API, the agent first researches middleware patterns, presents a plan with risk assessment, waits for approval, then implements and verifies the change. ## Quick Start Invoke the plan-execute workflow with a feature or refactoring goal, such as asking the agent to plan and then implement rate limiting for the API.
