explain.

Explain existing behavior through a concrete execution path. The worker traces and writes; the parent edits and presents. Explanation authorizes no product edits, and demonstrations stay within the user’s existing authorization.

Dispatch a capable worker before substantial investigation, honoring explicit user delegation and model choices; report the blocker if required delegation is unavailable. Answer directly only for an isolated single-fact lookup or when no worker can be dispatched, stating the reason first.

Give the worker the question unchanged, the audience/workspace/access context, and the trace rules below verbatim. Give each coherent flow one owner.

Trace

Give a working mental model of the asked behavior. Trace input to result, highlighting the decisions and failure behavior needed to understand it, with source locations for the key steps.

Ground the explanation in inspected code. Label execution results and inferred intent; name gaps that affect the answer. Return the answer and demonstration output inline; use one-off commands for focused demonstrations. Create files only for requested outputs or required project tooling.

Present

The parent edits for relevance, completeness, and clarity, and may spot-check a claim; the worker owns substantial tracing. Send concrete gaps back as bounded follow-up, and forward new evidence and user steering to the same owner.

Finish when the question is answered. If blocked, return what was traced, remaining unknowns, and the smallest missing input or access needed.