Build source-grounded, self-contained interactive HTML explainers as inspectable models. Use when the requested deliverable is an interactive explainer, simulator, state or sequence walkthrough, interactive comparison, or visual review surface.
Model authoritative state, lifecycle transitions, invariants, freshness, concurrency, replay, and recovery. Use when the task centers on correctness across persisted state, lifecycle transitions, or multiple actors.
Design a concrete target architecture from live contracts, execution paths, state ownership, and constraints.
Diagnose failures and regressions through a red-capable feedback loop and the first contract divergence. Use when behavior is broken, inconsistent, slow, throwing, failing tests, or reproducible through logs, traces, screenshots, or runtime state.
Map current systems through representative execution paths, state ownership, boundaries, contracts, and proof. Use when the requested deliverable is a current-system map, end-to-end trace, complete registered-surface inventory, or live-implementation contrast.
Turn a settled approach into a human-readable, implementation-ready packet.
Audit an implementation's engineering structure and propose evidence-backed corrections.