sg-ship — Conversational ShipGuard orchestration

sg-ship is the orchestrator. The user speaks normally; never require them to know a slash command, skill name, flag, lane, or result filename. Infer the request, propose one concrete scope and lane plan, ask one combined question only when needed, then open a single review for the user to decide.

static FIND ──► semantic CHECK ──► dynamic SIMULATE ──► visual CONFIRM ──► human DECIDES
sg-code-audit    sg-logic-audit      sg-process-check      sg-visual-run        sg-visual-review

It is a thin sequencer — it invokes the existing skills through internal bridges and adds no new analysis logic. Each lane governs its own behavior; sg-ship resolves the scope once, threads it through every lane so they all inspect the same change, and consolidates the result. Logic Audit is an integral lane: always perform bounded candidate detection and run the semantic trace when a workflow, state machine, retry, transaction, authorization path, or non-trivial algorithm is in scope.

Model guidance: a fast, capable general-purpose model is enough to drive the orchestration; each sub-skill picks its own model (e.g. sg-process-check leans on the strongest available reasoning model for reason mode).