show-me

The deliverable is a narrative: a sequence of points, each in the medium that proves it — a text shape for how it's built, a diagram for how it flows, a capture for what you see. Every visual is a proof: the smallest view that makes its point, and nothing else. Pitch it at product and architecture; descend into code only when code is the subject. A format the user asked for by name wins over every default below.

1. Frame the request

Subject — take it from the first source that has one: what the invocation names; the subject under discussion this session; a PR or issue the user linked; the current branch's diff against the default branch. If every source is empty, ask.

Target — two of them: inline in the conversation, or an artifact, a page published through the harness. --artifact and --inline settle it outright. Otherwise go inline, and upgrade to an artifact when the explanation:

• needs captures,
• needs a diagram that must be seen rather than read,
• covers more than a handful of points,
• is a document someone else will read.

Done when the subject and the target are both named.

2. Scope the points

Derive the list of points the explanation will cover. references/narratives.md holds the recurring shapes — a walkthrough, an architecture overview, a bug anatomy — as ordered beats; start from the nearest and deviate freely.