Explain Code

Answer a narrow question about the code with a small picture and a few sentences.
Catalogue and brevity rules are adapted from humanlayer's MIT-licensed show-me
skill; grounding and coverage disclosure are this repository's addition.

Deferred in v1: no HTML files, no writing any file, no browser.

Rules

1. Skip the preamble; keep prose brief (≤3 sentences beside the visual).
2. Pick the smallest catalogue form that makes the key point.
3. Include only the calls, files, and boundaries the current question needs.
4. Place each visual next to the short text it supports.
5. Every sketching reply ends with exactly one Grounding: line (see
   references/grounding.md). Clarification and whole-repo redirects do not.

Catalogue

Form	When	Reference
Call tree	Control flow / who calls whom	references/call-tree.md
Component tree	UI / module structure with paths	references/component-tree.md
File tree	Responsibility layout	references/file-tree.md
Sequence	Interaction over time (Mermaid)	references/sequence.md
Structural diff	What changed in shape	references/structural-diff.md

Grounding (summary)

Before a call tree, check freshness with
python3 "<skill-base-dir>/../refresh-architecture/scripts/run_architecture.py" --check
(exit 0 only = fresh). When fresh, obtain callers/callees from
python3 "<skill-base-dir>/../codebase-atlas/scripts/build_atlas.py" --tree ….
Never run --ensure or the analysis pipeline. Full contract, disclosure forms,
ungrounded reasons, and exemptions: references/grounding.md.

Whole-repository questions → name /codebase-atlas and stop (no Grounding: line).

Common Rationalizations

Rationalization	Why it's wrong
"I'll sketch from memory — the graph is probably fine"	Ungrounded trees look identical to grounded ones; run --check and disclose.
"A whole-repo atlas page answers a narrow question better"	That is /codebase-atlas's job; this skill stops and redirects.
"I'll refresh the graph myself so the answer is grounded"	The skill never runs --ensure or the pipeline; disclose ungrounded instead.

Red Flags

• A sketching reply with no final Grounding: line.
• A call tree that invents symbols not returned by --tree after a fresh check.
• The skill invoked --ensure, created or modified a file, or opened one for the user.

Verification

1. Confirm the catalogue form matches the question and cites a references/*.md.
2. Confirm sketching replies end with exactly one Grounding: line (or an exempt clarification/redirect).
3. Confirm sibling invocations use <skill-base-dir>/../codebase-atlas/ and <skill-base-dir>/../refresh-architecture/.