Learn From Code

Build codebase mental models and debugging intuition from code changes by prioritizing completeness of core mechanics over arbitrary length caps and 3-second visual comprehension: filter out peripheral boilerplate, explain every primary mechanism and guard with complete, unbroken causal reasoning, and visualize spatial math and state transitions with ASCII rulers, conversational vertical case walkthroughs, and semantic traffic-light icons (✅, ❌, ⚠️).

Operating Modes

• Mode A — Post-Implementation Review (Default): Explain a completed git diff, commit, or existing module using the Two-Turn Output Contract below, pausing after Section 5 for active retrieval before delivering Sections 6–8 in Turn 2.
• Mode B — In-Task Active Learning (TODO(human)): Survey existing patterns, generate surrounding boilerplate, and leave the core decision block marked TODO(human). Output a 4-bullet handoff block (Business Context — the domain rule at stake, Location — exact file & line of TODO(human), Contract — inputs and expected return/side effects, and Edge Cases to Weigh — boundary conditions to handle) and pause for the user to implement it; once written, review their implementation thoroughly and clearly.

Language & Developer Register Rules