buildChatGptUrl and its fitToBudget helpers have had no production
caller since prepareAction took over routing; the 2026-10-05 plan kept
them only for existing tests until then. They also shortened silently,
against the ask-before-shortening rule, so they were a trap to reuse.
Port the tests that guard live code onto applyTemplate and
serializePrompt, including the $&, $`, $' and $$ regressions (checked to
fail against the string-replacement form), and cover prepareAction's
whole-word cut, which nothing tested before.
Also drop getContext's inputType password check. menus.OnClickData has
no inputType, so it never fired; Firefox already withholds password
fields from selectionText via mozIsTextField(true) in SelectionUtils
(bug 565717).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The shortcut and toolbar read window.getSelection() without a limit, so a
select-all on a very large page handed prepareAction millions of
characters. Every binary-search probe re-encoded the whole candidate
prefix: about 1.1 s for 5M characters, paid again by the dialog.
Each code point adds at least one character to the URL, so nothing past
the first MAX_URL_CHARS can fit. Cap there before any encoding. A capped
source never takes the ready path: renderPrompt trims, so a prefix of
mostly whitespace could fit and silently drop the rest without consent.
Output is unchanged: 2,000 random cases matched develop exactly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
maxlength was hard-coded to 200 in the markup while MAX_MODEL_CHARS lives
in lib/defaults.js; lowering the constant would have left the input
accepting values mergeSettings() silently discards. It is now set from the
constant at render time.
Also records the hasFocus() gate's four manual frame cases in AGENTS.md,
including the findbar trade-off, and notes that fitToBudget()'s budget
guarantee assumes mergeSettings-validated input.