version: "1.0.1" name: show-me description: Explain the current conversation topic with the smallest visual — pseudocode, call tree, component/file tree, mermaid, or a diff. NOT for permanent architecture docs (use diagrams), interactive or narrated explainer artifacts (use create), or CSS/UI implementation (use visual). when_to_use: "explain this code, show me, draw this, diagram this, visualize this, sketch this, explain visually, show the call tree, show the component tree, quick diagram mid-conversation" user-invocable: true license: MIT metadata: author: HumanLayer (https://github.com/humanlayer/skills) homepage: https://github.com/humanlayer/skills/tree/main/plugins/show-me

Show Me

Explain the CURRENT conversation topic — not a permanent doc, not a generated artifact. ALWAYS honor the requested format; otherwise pick the smallest form that makes the point. ALWAYS route reusable interactive or narrated explainer artifacts to create, not this inline explanation skill.

Choosing the form

• Logic or an algorithm → pseudocode.
• Runtime control flow → an indented call tree, caller above callee.
• UI or module structure → a component tree annotated with the owning file

path.

• Cross-service or async interaction → a mermaid sequence/flow diagram.
• File layout or ownership → a shallow directory tree, one-line role per

entry.

• A change to any of the above → a diff-shaped block matching that same

shape (+/- lines), not prose describing the change.

• Layout, visual state comparison, or anything too dense for text/mermaid →

one local HTML file matching the project's real colors/type/spacing, opened with the available local browser or file opener.

ALWAYS

• ALWAYS pick one form, rarely two — never stack every form for the same

question.

• ALWAYS read the relevant code before explaining it; ALWAYS distinguish

observed behavior from assumptions.

• ALWAYS use real identifiers from the codebase (files, functions,

components) — never placeholder names.

• ALWAYS trim to only the calls/props/states/boundaries the current question

needs.

• ALWAYS keep prose to one or two sentences next to the visual — the visual

does the explaining.

• ALWAYS write the HTML fallback to a scratch path and open it locally.

NEVER

• NEVER publish the HTML fallback as a claude.ai Artifact — this repo is

local-output only (WISDOM § Documentation); ALWAYS use an available local opener on the file instead.

• NEVER force text or mermaid before a requested HTML explanation — ALWAYS

select the form by the question and requested format.

• NEVER let the visual outlive the conversation turn as a checked-in doc —

persistent diagrams belong to diagrams or specs, not here.

• NEVER build a multi-artboard deck or long-form asset here — that's

create's job.

Ported and adapted from humanlayer/skills show-me (MIT, © 2026 HumanLayer).