Tell every editor which key copies, since y there types a y

[hahwul]

^Y became Copy in every text box in #677. The footers only half followed: five INS strips named it, five did not.

The Fuzzer template editor was the pure case — it told you a ⇧arrow band could be built and then named no key that copies one. The bare y its READ strip advertises is, in that mode, a literal character that replaces the band you just built. One keystroke, no toast, selection gone. That is the bug report ^Y exists to answer, and the footer is where it lands.

The strips

The axis is a strip describing a mode it is not in.

Surface	Before	After
Fuzzer template INS	type · ⇧arrows select · ^Z undo · …	… select · **^Y copy** · ^Z undo · …
Repeater gRPC INS	type head/metadata · esc read · ↹ pane	… · **⇧arrows select · ^Y copy** · … · ↹ text
Rewriter preview INPUT	type sample HTTP · ↑ list · …	… · **⇧arrows select · ^Y copy** · …
JWT INPUT INS	type a JWT · esc read · …	… · **⇧arrows select · ^Y copy** · …
JWT HEADER / PAYLOAD	type JSON · … · space cmds · …	… · **⇧arrows select · ^Y copy** · …
JWT SECRET	type secret · … · space cmds · …	… · **^Y copy** · … (no band — single-line field)
Intercept held bytes	type to edit · ^R forward · …	… · **⇧arrows select · ^Y copy** · …

The Intercept editor named neither token, which is how it kept a registered ^Y that nothing on screen mentioned for its whole life.

On the JWT ENCODE panes and the Rewriter sample there is no READ mode at all, so ^Y is not the convenient copy — it is the only one. Those same strips advertised space cmds, which edit_json/edit_secret insert as a literal space; the menu that never opened was spending the one token with room to say which key copies.

Shown in INS throughout rather than only while a band is live — matching the five strips that already did, and #677's own decision not to gate the key on a selection. ^Y copy sits right after the select token so it survives an 80-column truncation.

What the footers promised, made true

A strip that names a band the copy ignores is worse than one that names nothing.

• jwt_copy consulted a band on one of its four editors. INPUT-in-INS copied the whole token while jwt_selection_active? reported the band as live — the split RepeaterView#pane_selection? documents in its own comment. HEADER/PAYLOAD were never asked. The four sibling tabs already agreed; JWT was the outlier.
• jwt_clear_selection cleared the READ model in a mode whose band lives on the editor, so the verb was a no-op exactly where this change makes bands copy-relevant.
• edit_json was the one printable arm of nine missing report_replaced — typing over a band on the pane this change teaches you to build one on lost text with no toast and no pointer at ^Z.
• rewriter_copy returned without a status on an empty pane — a dead key on a pane whose ^Y is its only copy.

↹ pane → ↹ text on three Repeater INSERT arms

editor_captures_tab? is request_text_editing?, which has no gRPC or sub-pane exclusion, so Tab splices a TAB into a header value. The plain-HTTP arm has said ↹ text since it grew its own INSERT branch; gRPC, split-decode and WS shared one string across both modes and kept the READ token.

Also retags ~24 comments and it descriptions still teaching ^Y for the chain editor, which moved to ^Q in #677 — now actively misleading, since ^Y is live.

Guard

spec/tui/editor_footer_copy_key_spec.cr, three source-grep rules:

1. a strip that advertises a band names a copy key;
2. a type … strip does not offer space cmds;
3. every tab whose Copy verb carries ^Y names it in a strip of its own.

Rule 3 is anchored to the keymap rather than a hand-kept list, and is the one that sees a strip saying neither token — the Intercept shape rule 1 is structurally blind to. All three were control-run against the pre-fix strings. Rule 3 passed on its first draft because a controller's own explanatory comment contains the token, so it now reads code lines only.

These are hand-written prose strings with no compiler behind them, and the pair drifted across ten strips.

Verification

• crystal spec — 8513 examples, 0 failures (the 8 capture_status/project_registry failures on this machine are a gori already holding :8070, identical before and after)
• crystal build --no-codegen clean; crystal tool format src spec clean; no new ameba findings
• Live in the TUI under tmux: Rewriter copied 3b on a 3-char band; JWT HEADER copied (6b) on a 6-char band where it would previously have said 15b; Fuzzer/JWT/Rewriter strips read as intended on screen