fix(people): find verified people who have no profile in New message, mentions and Command-K by morgmart · Pull Request #2668 · block/buzz-app

morgmart · 2026-11-05T20:47:49Z

Fixes AIDA-2449 and AIDA-2436.

What

People who have a relay-verified name but no profile could be added in Add members (#2554), but New message, @mentions and Command-K still couldn't find them.

This builds on #2554's rule instead of adding a second one:

No names are copied onto search results. The people directory returns someone found only by a verified name as an unnamed row. Each picker names it live through the existing verified-name lookup, which stays pinned to the relay's authority key. A head signed by anyone else finds nobody.
New message always searches typed names. A finished browse lists profiles only, so it can't prove that someone with only a verified name isn't there.
Mentions reuse a finished search, or treat an empty one as proof that nobody matches, only when the typed text is long enough for the relay to match verified names (a plain word of 3+ characters). They keep an unnamed row until its name is looked up.

Tests

Directory: finds a verified-only person, and ignores a head from another signer.
New message: finds a verified-only person by verified name and names the chip from it. Also covers searching after a finished browse.
Mentions: keeps an unnamed row on refinement, and searches again after a 2-character query.
The new tests fail on main and pass with this change.

Screenshots

No visual change. Rows use the existing pickers and labels.

🤖 Generated with Claude Code

kalvinnchau

🤖 Three verified-name search gaps remain: directory completion suppresses lookup, two-character mention caching suppresses refinement, and pending exact-key reads retain withdrawn authority names.

kalvinnchau

🤖 Three residual verified-name search defects and a page-completeness edge case remain. The prior three reported triggers are repaired.

kalvinnchau

🤖 Two remaining mention-name lifetime defects and a regression-coverage gap. The authority-withdrawal issue is an incomplete repair of the earlier finding; display-name matching, compound-prefix controls and raw-page pagination now pass the exercised cases.

kalvinnchau

🤖 The mounted chooser repair passes withdrawal/rotation checks. Remaining lifetime consumers and a caller regression gap are detailed inline.

… mentions and Command-K Builds on #2554's rule for Add members: a person the relay finds by a verified name alone comes back as an unnamed row, and the picker names it live from the community's verified lookup. No name is copied onto search results. - The people directory returns those rows, so New message, mentions and Command-K all see them. - New message always searches typed names: a finished browse lists profiles only and cannot rule them out. - Mentions reuse a finished search, or treat an empty one as proof of absence, only for text long enough for the relay to match verified names, and keep an unnamed row until its name is looked up. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

…reads The recipient search now checks verified names in the shared profiles. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

kalvinnchau

🤖 Three functional gaps remain in verified-name discovery/completion, plus a missing regression assertion. Details inline.

Every consumer of the people directory now names a person found by a verified name alone before offering or deciding on them: - Agent Share names rows like New message, and hides one still unnamed. - Space completion reads the verified names it needs before deciding, so a pending chip resolves even if the chooser never named the row. - Refining a cached mention page also matches the verified username. Adds a usePeople test that fails without the verified-name match keys. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

kalvinnchau

🤖 Two new pending-Space naming concerns. The namesake defect is reproduced in an isolated resolver test, but actual-app browser checks did not reproduce it; that discrepancy remains unresolved. The supersession concern is source-supported and lacks runtime confirmation.

Space completion had its own copy of the verified-name rules, and it drifted: it asked only about unnamed rows, missing a namesake with a profile, and could restore a name a newer shared head had withdrawn. The lookup that useVerifiedNames runs is now one function, lookUpVerifiedNames. The hook and the Space resolver both call it, and both read names back through verifiedName, so they share one set of authority and supersession rules. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

kalvinnchau

🤖 A verified-name read failure now allows pending Space resolution to select a namesake from incomplete evidence. Details inline.

…amed A person found by a verified name alone, whose name the lookup could not read, may be a namesake of anyone else on the page. Space returns to prose instead of picking from incomplete evidence, whether the read failed or simply did not answer. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

kalvinnchau

🤖 One remaining Space-completion consistency defect.

The mounted chooser and the pending resolver now share hasUnnamed: a page holding someone found by a verified name alone, whose name is not known yet, cannot decide Space in either path. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

kalvinnchau

🤖 One new merge-integration regression; the existing continuation finding remains unchanged.

hasUnnamed now takes the query: a pasted key names one person and has no namesakes, so its page always decides. The gate stays for names. Co-authored-by: Carl <c217fe6b9d958f41c3a5e030dccc7f626775a923089cb6491305eade75ea1f1b@buzz.block.builderlab.xyz> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: morgmart <98432065+morgmart@users.noreply.github.com>

morgmart marked this pull request as ready for review November 5, 2026 21:26

morgmart requested a review from a team as a code owner November 5, 2026 21:26

kalvinnchau reviewed Nov 5, 2026

View reviewed changes

Comment thread src/features/direct-messages/NewMessage.test.tsx Resolved

Comment thread src/bundled/mentions/useMentionDirectory.ts Resolved

Comment thread src/features/relay/direct-messages.ts Resolved

kalvinnchau reviewed Nov 5, 2026

View reviewed changes

Comment thread src/bundled/mentions/useMentionDirectory.ts Resolved

Comment thread src/features/direct-messages/usePeople.ts Resolved

Comment thread src/shared/search-debounce.ts Resolved

Comment thread src/features/relay/direct-messages.ts Resolved