Clone dialog repository metadata

Every row in the Clone dialog's GitHub repository list is a rich metadata card rather than a bare name, matching the Desktop Material v2 prototype. The card surfaces the information needed to pick the right repository without opening it in a browser first.

Behavior

• Each row shows the owner/name, a short description, and an inline metadata line with the primary language (a color dot plus the language name), stars, forks, on-disk size, the default branch, and a relative last-updated time.
• Metric icons come from the bundled Material Symbols set (star, fork_right, database, alt_route, schedule); the language dot uses a small fixed palette of GitHub Linguist-style colors with a neutral M3 outline fallback for any language not in the palette.
• A Public / Private pill (with a public or lock glyph) sits beside the name so visibility is explicit on every row, not only implied by an icon.
• Star and fork counts are formatted compactly (for example 4300 renders as 4.3k); size is scaled from the API's kilobytes into a human byte string (for example 8.1 MB); the timestamp renders as a relative "x ago" phrase.
• The filter panel adds language chips alongside the existing visibility chips. The chips are derived from the languages actually present in the loaded repository set, deduplicated case-insensitively and sorted, so the offered filters always reflect real data. Selecting languages is additive (multi-select); an empty selection means no language filter. Language chips are derived from the visibility-filtered set before the language filter is applied, so selecting one language never hides the other chips.

Data and configuration

• The GitHub listing type IAPIRepository carries the metadata additively: description, language, stargazers_count, forks_count, size, and updated_at. These fields flow straight from the /user/repos response; private and default_branch were already present.
• The fields are optional. Older GitHub Enterprise Server responses and the GitLab / Bitbucket adapters may omit any of them, in which case that single metric is dropped from the row while the rest still paint. A repository with no detected language contributes no chip and shows no language dot.
• The metadata card is opt-in per list via the showMetadata prop. The reused submodule and subtree pickers keep the compact single-line row.
• New visible labels (the visibility pill, the description fallback, the metric accessible names, and the language filter eyebrow) are localized in all three language modes — English, playful Hong Kong Cantonese, and the bilingual view — and update live when the language mode changes.

Performance and accessibility

The list stays virtualized. Each metadata card paints at a fixed 84px height (the description and metadata lines are single-line and height-pinned) so the list's pointer hit-testing matches the painted rows exactly. The metadata container is always rendered — even when a sparse response yields no metrics — so the row height never varies. Metric glyphs are decorative and hidden from assistive technology; each metric span carries an accessible label such as "4.3k stars" so screen readers announce the value with its meaning.

Failure modes and security

Metadata is presentational and derived only from the repository listing the account already has permission to read. Missing, null, or unparseable values (a blank description, an absent language, an unparseable timestamp) degrade to a neutral fallback or an omitted metric rather than an error. No additional network requests are made to populate the card.

Verification

Focused tests cover the language-color palette (known, case-insensitive, and neutral-fallback cases), language derivation and deduplication from a listing, compact count/size/relative-time formatting with graceful nulls, the additive metadata mapping onto list items, and the language filter's case-insensitive narrowing. Row-rendering tests assert the description, language dot and color, every metric, the Public/Private pill, the blank-description fallback, graceful omission of absent metrics, the English/Cantonese/bilingual labels, and that the compact (non-metadata) row still renders.