fix(web): enter activates highlighted search row (#406)

[intendednull]

what

caveman: enter on search input use to push query to recents pile, even
when row was highlighted. AT users see aria-activedescendant say row
1 selected, hit enter, nothing happen except recents grow. broken
since #344 / PR #403 added active_index but never wired enter to it.

fix

crates/web/src/components/search/input.rs enter branch:

• check active_index against flat results.len().
• in range -> fire new on_activate(SearchResult). same callback row
  click fires.
• out of range (no results, or stale index) -> fall back to
  on_submit(query) (recents push). empty-query affordance kept.
• always prevent_default so wrapping <form> no swallow the enter
  and tear surface down.

SearchSurface wires on_activate=on_select_result. keyboard and
mouse now converge on one navigation path.

tests

three new browser tests in
crates/web/tests/browser.rs mod phase_2e_search_enter_activate:

• enter_with_active_row_activates_that_row_not_recents_push - two
  rows, active_index=1, enter fires on_activate(m-1), submit silent.
• enter_with_no_results_falls_back_to_recents_push - empty results,
  enter fires on_submit(query), activate silent.
• enter_with_active_index_past_tail_falls_back_to_submit - stale
  index past tail, enter no panic, falls back to submit.

verify

local (no firefox/wasm-pack/geckodriver in worktree, CI runs headed
browser tests):

• cargo fmt --check clean
• cargo clippy --workspace --all-targets -- -D warnings clean
• cargo check -p willow-web --tests --target wasm32-unknown-unknown
  clean
• cargo test -p willow-web --lib 72 passed

wasm-pack browser run deferred to CI - geckodriver setup not in this
sandbox. expectation: the three new tests RED before patch (compile
failure - on_activate prop missing), GREEN after.

tradeoff

runner-up: keep one on_submit: Callback<String> and let surface
inspect active_index itself. rejected - couples surface to flat-order
walk that already lives in input.rs for arrow keys and
aria-activedescendant. one source of truth for "what does
active_index point at" stays inside the input component.

Refs #406

---

Generated by Claude Code

[intendednull]

What happened here? Why was this not merged?

[intendednull]

root cause: ci.yml + e2e.yml only trigger on pull_request: branches: [main]. PR #459 base = claude/adoring-euler-3axK7 (master-batch branch from /resolving-issues skill) → workflows never ran → no checks → coordinator stuck waiting for CI that can't fire.

skill's master-PR-batch pattern assumes sub-PRs trigger CI. this repo gates CI to main-only. pattern breaks here.

retargeting #459 base → main so CI runs. dropping master-batch coordination for this run — each sub-PR ships standalone into main, like #407/#408/#409 did before. will note in lessons-learned to either widen workflow triggers to claude/** branches or document the constraint in the skill.

---

Generated by Claude Code

[intendednull]

Fix the conflicts and make sure we have an issue open if the problem has not already been addressed in regards to the skill workflow.

[intendednull]

closing — duplicate of #433.

PR #433 (merged 2026-04-27 13:01) shipped same fix for #406: wire Enter to active_index, fall back to on_submit when no row, SearchInput on_select prop, browser tests. issue #406 closed automatically by maintainer the same day.

PR #459 dispatched 2026-04-28 04:19 by /resolving-issues skill that pulled the issue list before the dedup pass — issue showed state: open in the bulk list because of api caching / pagination timing. coordinator should have re-checked issue state before dispatching implementer; that's the second skill failure mode worth recording.

filing follow-up note in #465 (skill workflow tracking issue) — implementer agents should re-check issue state at start as a fast-fail guard.

closing PR + tearing down worktree. no code shipped.

---

Generated by Claude Code