fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation - #7337

…ion (block#7337)

## Problem

block#6732 added a busy-owner hold to the ACP harness: when a scope's
recorded session owner (`session_owners`) is checked out on **any**
turn, `dispatch_pending` holds the scope's batch instead of dispatching
it. The hold was added to keep one provider session per thread — but it
is unconditional: it applies to `Conversation` scopes too, and it has no
time bound.

Under the default `session_policy=channel`, every channel collapses to a
single `Conversation` scope, so once two channels' sessions land on the
same worker (pass 2 of `try_claim` picks the first idle worker by index,
so this happens quickly after any restart), channel A's mention starves
behind channel B's in-flight turn — for up to the full
`max_turn_duration` (7200s by default) — while other workers sit idle.
The only signal is a DEBUG-level log, and the 👀 seen-reaction is added
at queue admission *before* the hold decision, so the user sees the
agent acknowledge the mention and then nothing.

Observed in production on the first day of the v0.5.22 rollout: three
separate incidents where a mention got 👀 but no turn started until an
unrelated channel's turn ended on the shared worker (in the worst case
the blocking turn sat in a single tool call for 6+ minutes).

## Fix

One new seam, `AgentPool::hold_decision`, replaces the raw
`should_hold_for_busy_owner` check in `dispatch_pending` (the predicate
itself is unchanged and remains the inner check):

- **`Conversation` scopes never hold.** Channel-policy channels and all
DMs dispatch immediately; a busy owner means forking onto an idle
worker, exactly the pre-block#6732 behavior. This removes the cross-channel
head-of-line blocking entirely for the default policy.
- **`Thread` scopes hold for a bounded window.**
`HOLD_BUSY_OWNER_TIMEOUT` (10s) is measured from the first time the
batch is held (`held_since` stamp); once elapsed, the batch stops
holding and forks a fresh session on an idle worker, rebuilding thread
context from the relay. This preserves block#6732's session-continuity intent
for the momentary-busy case while capping the worst-case wait. No new
timer is needed: held batches are requeued with preserved timestamps and
re-evaluated on every dispatch trigger (turn end, relay event, 30s
maintenance tick), so the effective worst-case re-check gap on a fully
silent system is one maintenance tick.
- **Holds are observable.** Holding logs at INFO and a hold expiry logs
at WARN (previously DEBUG-only), and both emit observer-feed events
(`busy_owner_hold`, `busy_owner_hold_forked`) with the scope, owner
index, and held duration.

`held_since` is derived state and is cleared on every removal path:
dispatch/fork (inside `hold_decision`), `invalidate_channel_sessions`,
`invalidate_scope_session`, and `switch_idle_agent_model`.

## Accepted trade-offs

- A fork after an expired hold leaves the old owner's now-orphaned
thread session in its session map until natural rotation/invalidation —
benign, and identical to pre-block#6732 fork semantics (`loadSession: false`;
sessions are worker-pinned, so migration is not an option).
- Under sustained pool exhaustion the hold stamp is cleared on the fork
attempt and re-stamped next cycle, so the bound is effectively "timeout
after a worker frees up," not absolute wall clock.

## Tests

- New table test `hold_decision_covers_variant_session_busy_and_timeout`
over the full input space (scope variant × idle-session presence × owner
busyness × elapsed vs. window). The `Conversation` + busy-owner row is
the cross-channel regression guard; the past-window row guards the
bound. Both were mutation-checked: removing the variant gate or the
timeout branch fails the suite.
- `busy_session_owner_holds_batch_instead_of_forking_session` extended
with the Hold → ForkAfterHold transition, the `Conversation` dispatch
guard, and `held_since` pruning on channel invalidation.
- Scope-invalidation and idle-model-switch tests extended to cover
`held_since` cleanup alongside the existing `session_owners` assertions.

Signed-off-by: Will Pfleger <[email protected]>
(cherry picked from commit b17c077)
Signed-off-by: nambse <[email protected]>

* origin/main:
  fix(desktop): keep packaged frontendDist relative so Windows embeds assets (#7177)
  fix(sidebar): simplify unread indicators and emphasize priority activity (#7134)
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  chore(release): release Buzz Desktop version 0.5.22 (#7308)
  feat(desktop): preserve mentions across copy and paste (#7228)
  test(desktop): await Bestie drag and profile hover endpoints (#7294)

Signed-off-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>

* origin/main:
  chore(release): release Buzz Desktop version 0.5.23 (#7381)
  fix(desktop): keep packaged frontendDist relative so Windows embeds assets (#7177)
  fix(sidebar): simplify unread indicators and emphasize priority activity (#7134)
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  chore(release): release Buzz Desktop version 0.5.22 (#7308)
  feat(desktop): preserve mentions across copy and paste (#7228)
  test(desktop): await Bestie drag and profile hover endpoints (#7294)

Signed-off-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>

…-enforcement

* origin/main:
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  chore(release): release Buzz Desktop version 0.5.22 (#7308)
  feat(desktop): preserve mentions across copy and paste (#7228)
  test(desktop): await Bestie drag and profile hover endpoints (#7294)
  Collapse contiguous join messages (#7262)
  chore(release): release Buzz Desktop version 0.5.21 (#7301)
  fix(scripts): copy global-agent-config.json in buzz-adopt-prod-agents (#7303)

Signed-off-by: Hayt <9e1c23a3fd83f61da34420e4e88ff1b16e45cafcc0cd9019eb07d4ecfa8ca9b0@buzz.block.builderlab.xyz>

…n-surface

* origin/main: (23 commits)
  chore(release): release Buzz Desktop version 0.5.23 (#7381)
  fix(desktop): keep packaged frontendDist relative so Windows embeds assets (#7177)
  fix(sidebar): simplify unread indicators and emphasize priority activity (#7134)
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  chore(release): release Buzz Desktop version 0.5.22 (#7308)
  feat(desktop): preserve mentions across copy and paste (#7228)
  test(desktop): await Bestie drag and profile hover endpoints (#7294)
  Collapse contiguous join messages (#7262)
  chore(release): release Buzz Desktop version 0.5.21 (#7301)
  fix(scripts): copy global-agent-config.json in buzz-adopt-prod-agents (#7303)
  ...

Signed-off-by: Will Pfleger <[email protected]>

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[#7337](#7337 (review)).

### Related issue

Follow-up to #7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

…#7340)

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[block#7337](block#7337 (review)).

### Related issue

Follow-up to block#7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>
Signed-off-by: Trevor P <[email protected]>

* fix(mobile): style inline code with the app mono face (block#6631)

## Summary

Inline code on mobile renders as **bold body text on a faint background
wash** — no monospace face, no chip, and it cannot wrap. block#5257 diagnosed
this as a missing `highlightBuilder`.

That is no longer the right fix. `gpt_markdown` 1.2.0 deprecates
`highlightBuilder` (removal in 2.0.0), renders inline code as a real
chip, and adds `InlineCodeStyle` for restyling it. The package author
confirmed this on the issue. So this PR is an upgrade — 1.1.6 → 1.2.1 —
plus one theme declaration, rather than the builder the issue originally
asked for.

**Where the style is declared.** `GptMarkdownThemeData` goes in
`AppTheme._buildTheme`, which both `light()` and `dark()` call. That
reaches all four `GptMarkdown` call sites — `message_content`,
`transcript_item_widget`, `token_pill`, `custom_emoji_render` — so the
style is stated once instead of per widget. A widget-level
`inlineCodeStyle` would have covered channel messages only, leaving the
other three on the package's defaults.

**What is declared.** Face, size, ink, chip fill and outline — not the
face alone. A face name on its own leaves the rest on the package's
defaults, which put inline code at 14.1sp beside a fenced block's 13, on
a neutral `onSurface` tint rather than the app's code surface. In dark
that tint is *lighter* than the surface, while every other code surface
in the app is recessed, so the chip read as a different kind of object.
All of it now comes from one `CodeStyle` declaration that the fenced
block reads from too, so the two cannot be edited apart.

**Three adaptations the upgrade requires.** Each was found by running
the gate, not by reading the changelog:

1. **`imageBuilder` widened** to `(context, url, width, height)`. This
is a hard compile error, and it is **not listed in the package's
migration guide**, which states "nothing here stops code compiling".
Worth reporting upstream.
2. **`autolink` now defaults to `true`.** `normalizeBareLinks()` already
rewrites bare URLs into Markdown links before rendering, so both would
run. `message_content` opts out with `autolink: false` to keep current
behaviour exactly. The migration guide argues for dropping the
pre-processor instead — a better fix, but a behavioural change that
belongs in its own PR.
3. **`gpt_markdown.dart` now re-exports `markdown_config.dart`**, making
two direct imports redundant. `flutter analyze` reports `No issues
found!` on 1.1.6 and flags both on 1.2.1, so these warnings are new, not
pre-existing.

**Deliberately out of scope.** The three non-message call sites now
autolink bare URLs, since only `message_content` has a pre-processor to
collide with. Custom inline components (`_MentionMd`, `CustomEmojiMd`,
`_ChannelLinkMd`) could additionally declare `allScopesExceptLinkLabel`
— 1.2.0 offers it as the fix for a `WidgetSpan` chip going blank inside
a link label on iOS — but current behaviour is unchanged without it, so
that stays a separate change.

### Related issue

Fixes block#5257

Duplicate scan: searched `gpt_markdown`, `inline code mobile`,
`highlightBuilder` and `InlineCodeStyle` across both PRs and issues. No
open PR touches inline code styling. block#6135 (link labels) and block#6166 (text
selection) also touch mobile Markdown but address different defects.

### Testing

Full gate, `just ci` — exit 0:

| Stage | Result |
|---|---|
| Rust (33 suites) | 4768 passed, 0 failed |
| Desktop | 5799 passed, 0 failed |
| Mobile | **2011 passed**, 0 failed |
| `flutter analyze` | `No issues found!` |
| Desktop + web build | ok |

Run on the branch with `main` merged in, so these numbers match what CI
builds.

**New regression test** — `renders inline code in the app code style`.
It resolves the `CodeTextSpan` the package tags inline code with, which
carries both the resolved `TextStyle` and the colours the chip behind it
is painted with, so face, size, ink, fill and outline are all asserted
rather than a widget's presence. It is negative-controlled: reverting
only the theme declaration fails it with

```text
Expected: a numeric value within <0.001> of <13.0>
  Actual: <14.1>
```

and dropping the declaration entirely falls back to
`packages/gpt_markdown/JetBrainsMono` — so the test measures the real
thing, and it would catch a future regression that silently drops the
theme extension.

The test passes `baseStyle: messageBodyTextStyle`, the style the message
surfaces actually use; the widget's own fallback is the smaller
`bodyMedium`, which would move the expected size.

The test finds paragraphs with `find.byWidgetPredicate((widget) =>
widget is RichText)`, not `find.byType(RichText)`: inline code renders
through `BidiRichText`, a `RichText` subclass, and `byType` matches
exact runtime types.

That is a hazard for any test that reads text back out of a paragraph,
and one landed after this branch was cut:
`message_content_custom_emoji_test.dart` arrived with block#6996 and its
`code keeps literal emoji while adjacent known tokens render` case reads
a code span through `find.byType(RichText)`. It passes on `main` and
fails on the merge result, which is what CI builds, so it went red only
once CI was authorized. It now uses the same predicate. The two other
`byType(RichText)` call sites — the rest of that file and
`message_author_meta_test.dart` — were re-run and pass: their content
carries no code span, so the exact type still matches. They were left
alone.

### Screenshots

Rendered through the real `MessageContent` widget with the app's own
fonts loaded, at 390pt wide, 3x DPR. Sample text: ``Set `BUZZ_RELAY_URL`
before launch, then run `just mobile-test` to verify.``

| | Before (1.1.6) | After (1.2.1) |
|---|---|---|
| Light |
![before-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-light.png)
|
![after-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-light.png)
|
| Dark |
![before-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-dark.png)
|
![after-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-dark.png)
|

Before: bold Inter on a flat wash, no chip edge, and `just mobile-test`
breaks across the line with the wash simply ending. After: Geist Mono in
a bordered, rounded chip, and the wrapped fragment gets its own chip on
each line.

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>

* fix(buzz-acp): wake held ACP threads and fence forked sessions (block#7340)

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[block#7337](block#7337 (review)).

### Related issue

Follow-up to block#7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Signed-off-by: Salman Mohammed <[email protected]>
Co-authored-by: TolgaCinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>
Co-authored-by: Salman Mohammed <[email protected]>

* origin/main: (29 commits)
  fix(acp): pace targeted overflow recovery on consumer capacity (#7325)
  fix(link-preview): keep composer fetches user-paced (#7211)
  feat(mesh): upgrade to mesh-llm 0.76.0-rc8 and recommend Qwen3.8 27B (#6189)
  fix(agent): route GPT-5+ model-service FQNs to Responses (#7358)
  fix(buzz-acp): wake held ACP threads and fence forked sessions (#7340)
  fix(mobile): style inline code with the app mono face (#6631)
  chore(release): release Buzz Desktop version 0.5.23 (#7381)
  fix(desktop): keep packaged frontendDist relative so Windows embeds assets (#7177)
  fix(sidebar): simplify unread indicators and emphasize priority activity (#7134)
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  ...

Signed-off-by: Tom Brow <[email protected]>

* origin/main: (77 commits)
  fix(acp): pace targeted overflow recovery on consumer capacity (#7325)
  fix(link-preview): keep composer fetches user-paced (#7211)
  feat(mesh): upgrade to mesh-llm 0.76.0-rc8 and recommend Qwen3.8 27B (#6189)
  fix(agent): route GPT-5+ model-service FQNs to Responses (#7358)
  fix(buzz-acp): wake held ACP threads and fence forked sessions (#7340)
  fix(mobile): style inline code with the app mono face (#6631)
  chore(release): release Buzz Desktop version 0.5.23 (#7381)
  fix(desktop): keep packaged frontendDist relative so Windows embeds assets (#7177)
  fix(sidebar): simplify unread indicators and emphasize priority activity (#7134)
  Add generic information-flow control core (#7293)
  feat(buzz-acp): update base prompt; add buzz context and skills to Pi agents (#7335)
  fix(desktop): restore mention chip identity icons (#7338)
  Persist video playback speed preference (#7336)
  Verify ACP relay events before prompt routing (#7010)
  fix(buzz-acp): bound busy-owner hold to prevent cross-channel starvation (#7337)
  feat(desktop): invite owned agents from standalone forums (#7125)
  fix(desktop): authorize remote mentions at publication (#7124)
  fix(acp): rename system tag to agent-instructions (#7332)
  fix(desktop): bind duplicate mention selections to exact recipients (#7133)
  refactor(relay): extract NIP-29 membership authorization (#7285)
  ...

Signed-off-by: Sol <478bb5a31222ea2b28a3d1afb8b1d598940628f19c2a87efc3c4b822299eeec6@buzz.block.builderlab.xyz>

# Conflicts:
#	desktop/src-tauri/src/commands/media_download.rs
#	desktop/src-tauri/src/lib.rs

* fix(mobile): style inline code with the app mono face (#6631)

## Summary

Inline code on mobile renders as **bold body text on a faint background
wash** — no monospace face, no chip, and it cannot wrap. #5257 diagnosed
this as a missing `highlightBuilder`.

That is no longer the right fix. `gpt_markdown` 1.2.0 deprecates
`highlightBuilder` (removal in 2.0.0), renders inline code as a real
chip, and adds `InlineCodeStyle` for restyling it. The package author
confirmed this on the issue. So this PR is an upgrade — 1.1.6 → 1.2.1 —
plus one theme declaration, rather than the builder the issue originally
asked for.

**Where the style is declared.** `GptMarkdownThemeData` goes in
`AppTheme._buildTheme`, which both `light()` and `dark()` call. That
reaches all four `GptMarkdown` call sites — `message_content`,
`transcript_item_widget`, `token_pill`, `custom_emoji_render` — so the
style is stated once instead of per widget. A widget-level
`inlineCodeStyle` would have covered channel messages only, leaving the
other three on the package's defaults.

**What is declared.** Face, size, ink, chip fill and outline — not the
face alone. A face name on its own leaves the rest on the package's
defaults, which put inline code at 14.1sp beside a fenced block's 13, on
a neutral `onSurface` tint rather than the app's code surface. In dark
that tint is *lighter* than the surface, while every other code surface
in the app is recessed, so the chip read as a different kind of object.
All of it now comes from one `CodeStyle` declaration that the fenced
block reads from too, so the two cannot be edited apart.

**Three adaptations the upgrade requires.** Each was found by running
the gate, not by reading the changelog:

1. **`imageBuilder` widened** to `(context, url, width, height)`. This
is a hard compile error, and it is **not listed in the package's
migration guide**, which states "nothing here stops code compiling".
Worth reporting upstream.
2. **`autolink` now defaults to `true`.** `normalizeBareLinks()` already
rewrites bare URLs into Markdown links before rendering, so both would
run. `message_content` opts out with `autolink: false` to keep current
behaviour exactly. The migration guide argues for dropping the
pre-processor instead — a better fix, but a behavioural change that
belongs in its own PR.
3. **`gpt_markdown.dart` now re-exports `markdown_config.dart`**, making
two direct imports redundant. `flutter analyze` reports `No issues
found!` on 1.1.6 and flags both on 1.2.1, so these warnings are new, not
pre-existing.

**Deliberately out of scope.** The three non-message call sites now
autolink bare URLs, since only `message_content` has a pre-processor to
collide with. Custom inline components (`_MentionMd`, `CustomEmojiMd`,
`_ChannelLinkMd`) could additionally declare `allScopesExceptLinkLabel`
— 1.2.0 offers it as the fix for a `WidgetSpan` chip going blank inside
a link label on iOS — but current behaviour is unchanged without it, so
that stays a separate change.

### Related issue

Fixes #5257

Duplicate scan: searched `gpt_markdown`, `inline code mobile`,
`highlightBuilder` and `InlineCodeStyle` across both PRs and issues. No
open PR touches inline code styling. #6135 (link labels) and #6166 (text
selection) also touch mobile Markdown but address different defects.

### Testing

Full gate, `just ci` — exit 0:

| Stage | Result |
|---|---|
| Rust (33 suites) | 4768 passed, 0 failed |
| Desktop | 5799 passed, 0 failed |
| Mobile | **2011 passed**, 0 failed |
| `flutter analyze` | `No issues found!` |
| Desktop + web build | ok |

Run on the branch with `main` merged in, so these numbers match what CI
builds.

**New regression test** — `renders inline code in the app code style`.
It resolves the `CodeTextSpan` the package tags inline code with, which
carries both the resolved `TextStyle` and the colours the chip behind it
is painted with, so face, size, ink, fill and outline are all asserted
rather than a widget's presence. It is negative-controlled: reverting
only the theme declaration fails it with

```text
Expected: a numeric value within <0.001> of <13.0>
  Actual: <14.1>
```

and dropping the declaration entirely falls back to
`packages/gpt_markdown/JetBrainsMono` — so the test measures the real
thing, and it would catch a future regression that silently drops the
theme extension.

The test passes `baseStyle: messageBodyTextStyle`, the style the message
surfaces actually use; the widget's own fallback is the smaller
`bodyMedium`, which would move the expected size.

The test finds paragraphs with `find.byWidgetPredicate((widget) =>
widget is RichText)`, not `find.byType(RichText)`: inline code renders
through `BidiRichText`, a `RichText` subclass, and `byType` matches
exact runtime types.

That is a hazard for any test that reads text back out of a paragraph,
and one landed after this branch was cut:
`message_content_custom_emoji_test.dart` arrived with #6996 and its
`code keeps literal emoji while adjacent known tokens render` case reads
a code span through `find.byType(RichText)`. It passes on `main` and
fails on the merge result, which is what CI builds, so it went red only
once CI was authorized. It now uses the same predicate. The two other
`byType(RichText)` call sites — the rest of that file and
`message_author_meta_test.dart` — were re-run and pass: their content
carries no code span, so the exact type still matches. They were left
alone.

### Screenshots

Rendered through the real `MessageContent` widget with the app's own
fonts loaded, at 390pt wide, 3x DPR. Sample text: ``Set `BUZZ_RELAY_URL`
before launch, then run `just mobile-test` to verify.``

| | Before (1.1.6) | After (1.2.1) |
|---|---|---|
| Light |
![before-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-light.png)
|
![after-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-light.png)
|
| Dark |
![before-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-dark.png)
|
![after-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-dark.png)
|

Before: bold Inter on a flat wash, no chip edge, and `just mobile-test`
breaks across the line with the wash simply ending. After: Geist Mono in
a bordered, rounded chip, and the wrapped fragment gets its own chip on
each line.

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>

* fix(buzz-acp): wake held ACP threads and fence forked sessions (#7340)

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[#7337](https://github.com/block/buzz/pull/7337#pullrequestreview-5116329341).

### Related issue

Follow-up to #7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

* fix(agent): route GPT-5+ model-service FQNs to Responses (#7358)

## Summary
Route Databricks Unity Catalog model services to OpenAI Responses when
their service name matches GPT-5 or newer. These models can reject tools
plus reasoning on Chat Completions.

Match only the service component, using the existing family-token
boundaries and a numeric major version. Catalog and schema names cannot
select the protocol. Keep neutral effort capabilities and the full model
ID unchanged; other services still use MLflow Chat Completions.

Keep the Rust and desktop resolvers in sync, add shared boundary cases
and a captured-HTTP regression for completion and summarization, and
update the documented FQN rule.

### Related issue
No duplicate found in searches for “FQN responses” PRs or “astra”
issues. Related: #6918 introduced Unity Catalog discovery.

Originating conversation:
buzz://message?channel=0b881928-a3a6-4c01-b981-8e64268f01ce&id=770949343bc96a9ed88acd90a1b37d358a0efc52c79237d0fdb491ce02b8d4ed

### Testing
No live Databricks inference test. The gateway must accept the full
model-service ID on its OpenAI Responses route; this remains the
integration risk.

The local `just ci` attempt exceeded its five-minute deadline during
`mobile-check`, so the full repository gate was not completed. All
push-hook checks passed.

Generated with Codex

Signed-off-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>
Co-authored-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>

* feat(mesh): upgrade to mesh-llm 0.76.0-rc8 and recommend Qwen3.8 27B (#6189)

Upgrades Buzz's mesh-llm dependencies through the released `v0.76.0-rc8`
tag (`2040765d`), including the Qwen3.8 curated recommendation and rc8's
scheduler/runtime improvements.

**Scope note:** the earlier open-relay/unenforced-admission mode has
been removed from this PR at Mic's direction — it is not a product mode
we want. Mesh admission remains roster/allowlist driven, exactly as on
`main`: on a relay with no NIP-43 membership snapshot the mesh runs
self-only. No NIP-11 mode probing, no mode-transition restarts. A future
perimeter/admission strategy for open relays will be designed
separately.

This PR also:
- seeds `BUZZ_AGENT_LLM_TIMEOUT_SECS=660` for mesh agents, above
MeshLLM's 600-second backend timeout;
- makes `desktop-tauri-clippy` lint both default and `mesh-llm` cfg
graphs;
- runs the feature-enabled desktop test suite in CI;
- recommends Qwen3.8 27B Q4_K_M for 64 GB-and-larger machines, then
ladders down through Gemma 4 E4B and Qwen 9B for smaller machines;
- keeps stored shared-compute `auto` translated to MeshLLM's supported
wire model `mesh`.

RC8 verification:
- `just ci` passed locally at
`92ecc7ec933bdd4df804cc9f28a2b51efa5313c5`.
- Pre-push differential gates passed, including both desktop Tauri cfg
graphs and package tests.
- A prior isolated runtime smoke used the RC8 binary's OpenAI endpoint
for a Buzz-shaped system/user/tool/tool-result/final-response loop; all
assertions passed and the isolated process was shut down.

Perf previously measured on M5 Metal, Qwen3.8-27B-Q4_K_M: TTFT 0.22–0.32
s, ~25 tok/s streaming; agent-shaped turns ~1 s to first token after the
first (prefix cache).

---------

Signed-off-by: Michael Neale <[email protected]>
Signed-off-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Signed-off-by: Alessandro Joabar <[email protected]>
Co-authored-by: Michael Neale <[email protected]>
Co-authored-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Co-authored-by: Mic Neale <[email protected]>
Co-authored-by: Alessandro Joabar <[email protected]>

* fix(link-preview): keep composer fetches user-paced (#7211)

**Category:** fix
**User Impact:** Link previews can keep loading while a message is being
composed, while sending still has a finite escape hatch and stalled
network transports cannot occupy preview slots forever.

**Problem:** Native metadata and image deadlines could collapse slow
previews into fallback cards while the user was still composing, and a
shared image-host cooldown made pasted batches fail inconsistently after
one rate limit. **Solution:** Keep preview resolution user-paced with no
aggregate request deadline, bound transport inactivity (15s DNS/connect,
30s idle read), serialize image requests by host, and allow at most one
server-directed cooldown wait of up to 30s across an image fetch and its
redirects. The existing bounded post-Send preparation and immediate Skip
paths remain unchanged.

<details>
<summary>File changes</summary>

**desktop/src-tauri/src/commands/link_preview.rs**
Removes aggregate native deadlines so composer metadata work can
complete at the user's pace, while retaining DNS/connect/idle-read
liveness bounds. Adds bounded host-paced image request coordination that
releases its gate during cooldown, waits inline at most once for at most
30 seconds, and cannot renew that wait through redirects or the outer
transient retry. Same-host image and favicon requests remain
deliberately serialized to align with host rate limits.

**desktop/src-tauri/src/commands/link_preview_rate_limit.rs**
Adds a fixed-size striped host gate so concurrent image requests are
serialized without retaining an unbounded attacker-controlled hostname
map.

**desktop/src-tauri/src/commands/link_preview_tests.rs**
Moves native link-preview tests into a dedicated module and covers the
user-paced metadata contract, bounded one-shot cooldown behavior, and
gate release while a rate-limited request sleeps—including a different
host sharing the same bounded gate stripe.

**desktop/src-tauri/src/commands/link_preview_youtube.rs**
Removes the thumbnail fetch deadline so YouTube previews follow the same
composer lifecycle contract while using the shared bounded transport.

**desktop/src/shared/lib/useResolvedLinkPreviews.ts**
Adds development-only metadata outcome diagnostics with elapsed time and
image/fallback state, without logging encoded image payloads.

</details>

### Reproduction steps

1. Open the desktop composer and paste several GitHub pull request links
whose OpenGraph images share a host.
2. Observe that image requests are paced by host instead of racing, and
slow-but-progressing preview work remains pending rather than
immediately becoming a completed favicon fallback.
3. Send while preview work is still pending and confirm **Preparing link
preview** remains bounded by the existing post-Send budget.
4. Use **Skip** during preparation and confirm the message proceeds
immediately.
5. In a development build, inspect the console for `[link-preview]
metadata fetch completed` diagnostics containing elapsed time and image
state without base64 payloads.

### Related issue

N/A — scoped from the linked Buzz implementation room.

### Testing

At current head `dfb394aafbee537e9ffb04ad3732d08f65f30b8e`:

- Production-bound paused-time metadata regression passed through
`fetch_link_preview_metadata`; restoring the former 10-second aggregate
wrapper makes it fail at the pending assertion.
- Native link-preview module: 19/19 passed.
- `cargo check --manifest-path desktop/src-tauri/Cargo.toml` passed.
- Rust formatting and `git diff --check` passed.
- Pre-push `push-head-scope`, org safety, differential file-size,
branch-skew, and `desktop-tauri-checks` hooks passed.

At prior head `59e2dcf167b15c7a3e637ad2608008b7f9cef5f3`:

- Full Tauri Rust suite: 3,056 passed, 19 ignored; integration crates 7
+ 3 passed.
- Focused native link-preview suite: 26/26 passed.
- The pasted multi-preview workflow was exercised in the desktop app and
confirmed improved before draft publication.

---------

Signed-off-by: Taylor Ho <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@users.noreply.github.com>

* fix(acp): pace targeted overflow recovery on consumer capacity (#7325)

🤖
## Summary

When a Buzz agent falls behind on incoming messages, its connection can
make the backlog worse while trying to recover. The connection buffers
messages from the relay server until the agent is ready to process them;
if that buffer overflows, recovery previously requested history for
**every subscribed channel** and paused socket reads while sending those
requests. That adds traffic to an already overloaded connection. This
change requests history only for affected subscriptions, once the code
consuming those messages has room, with at least five seconds between
attempts.

The recovery path now:

- Combines repeated losses into one pending recovery per affected
subscription, keeping the oldest dropped timestamp so replay starts
early enough.
- Waits until at least half the consumer queue is free and the relay's
existing rate-limit delay has expired. The queue wakes recovery when
space becomes available; recovery does not periodically sample capacity
or hold queue space away from live messages.
- Attempts one subscription at a time, choosing the least recently
attempted so a busy channel cannot crowd out other channels or
membership notifications. The five-second delay starts when an attempt
finishes, including a failed write; failed writes leave recovery
pending.

Recovery is paced by available capacity, not by how often messages are
lost. This is not a larger buffer or a cutoff that abandons recovery.
Subscription identifiers, message filters, replay timestamp overlap and
duplicate filtering are unchanged; no downstream agent changes are
required.

This targets a reproducible overload **amplifier**, not every cause of
overload or every catch-up limitation. The initial live overload's cause
has not been established. Recovery remains best effort: a successful
request write is not proof of delivery, and existing history/retention
limits, bounded duplicate tracking and replay limitations still apply.
There is no exactly-once or complete catch-up guarantee. A stalled write
can still pause socket reads for the existing ten-second timeout; the
pacing bound does not cover initial subscriptions, reconnects or other
retry paths.

### Related issue

Closest related: #5014 (channel re-subscription); also #6661 (membership
reconciliation) and #6090 (relay backpressure gap signaling). This
addresses local overflow recovery scheduling, not those separate
mechanisms.

### Testing

Recorded offline comparisons against the previous behavior, with the
final implementation at `8000636f3073167c5a5107bb179c7d91160f1729`:

| Same fixture: 18 subscriptions, three overload rounds | Before | After
|
| --- | --- | --- |
| Recovery history requests | 108 | 3 |
| Ping-response delay | About 4.6 seconds | Below the measurement's 1 ms
resolution |

A separate bounded-history fixture delivered all 320 events plus
subsequent live traffic in **both** versions. Regression coverage
exercises the real socket-handling task, including intermittent consumer
capacity, fairness, failed writes and cancellation of capacity waits
before live delivery. These are synthetic results, not production
throughput measurements or evidence of a deployed cure.

The full local `RUST_TEST_THREADS=4 just ci` run passed on September 4,
2026. Earlier unsuccessful local runs remain part of the validation
history. The [recorded validation evidence and separate desktop
follow-up](https://github.com/block/buzz/pull/7325#issuecomment-5540592398)
preserve the original desktop mock-history scroll failure, its passing
rerun and the remaining investigation. That desktop path does not run
the agent connection code; neither this repair nor the passing rerun
fixes the observed scroll problem.

---------

Signed-off-by: Logan Johnson <[email protected]>

* fix(mobile): render push notification sender identity as npub (#7494)

🤖

## Summary

When an iOS push notification comes from someone the app has no cached
name for, the notification title showed the first characters of the
sender's raw public key — for example `aa4fc866…`. That fragment is
unreadable and doesn't match how the same person appears anywhere else
in Buzz. This PR changes that title to the compact form of the sender's
npub (npub is the human-readable encoding of a Nostr public key): first
8 and last 4 characters — for example `npub14f8…9nsy`, the same identity
shape used across the desktop and mobile apps.

- Unnamed senders: raw hex fragment → compact npub.
- Named senders: unchanged — a sender the app has a display name for
still titles the notification with that name.
- Unverifiable sender identities (malformed keys, or lookalike strings
that are not literal 64-hex-digit keys) now render a neutral "Someone"
instead of partial raw key material.
- Everything else about the notification is unchanged: body text,
subtitle, thread matching and grouping, deep-link navigation, thread
identifiers, and the internal hex public key the resolver matches on.

The native iOS notification-service package (`BuzzPushKit`) gains a
minimal in-house bech32 codec (bech32 is the checksummed string encoding
npubs use) — checksum-validated, 32-byte keys only, and no new external
dependency. The hex input branch accepts exactly a 64 ASCII hex digit
key before any parsing, so strings that merely parse like hex (for
example a run of `+a` pairs) cannot become a displayed identity; this is
input validation for presentation. Event signature verification is
untouched.

### Related issue

Fixes: N/A. Searched existing issues/PRs for push-notification npub
identity — closest related: none found.

### Testing

At head `3e3f2813b8864b76257ccb50dea3a4b31fa4de0d` (base
`44316ff72f5f7de014c66b01cbf534298a70c249`; 4 files, +321/−4):

- CI `Mobile Swift` lane, at this exact head — all passed: `swift test`
(73 tests, 0 failures), the SwiftPM debug and release builds of
`mobile/ios/BuzzPushKit`, and the unsigned iOS release build.
- Test coverage: npub encoding cross-checked against independent
nostr-rs/NIP-19 vectors; rejection of bad checksums, mixed case, wrong
lengths, invalid alphabet, padding, and non-32-byte payloads; resolver
boundary matrix — hex/npub/invalid sender keys render compact npub or
"Someone" while body, subtitle, sender key, and thread identifier pass
through; named senders keep cached display names.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): shared npub identity foundation (canonicalNpub, PubKey gate, strict parser) (#7488)

🤖
## Summary

Identity keys in the desktop app are displayed as raw 64-character hex.
A person's key shows up as something like `953d3363…` — unreadable,
impossible to recognize as the same identity on another screen, and a
hazard when copied by hand. Nostr (the protocol Buzz runs on) has a
human-readable spelling for identity keys — the `npub1…` form — but the
desktop app did not use it consistently.

This is the foundation of the desktop npub changes: it adds the shared
pieces every identity surface builds on, and two follow-up slices stack
directly on this branch — #7489 converts the identity controls (profile,
settings, allowlist, workflow key fields) and #7495 converts the
everyday display surfaces (mentions, member lists, sidebar, and other
name fallbacks).

After this change:

- The shared identity widget shows the compact npub form —
`npub1j57...fjmv` — instead of a hex prefix, everywhere it renders (for
example the owned-agent public-key row on a profile). Copying it puts
the full npub on the clipboard.
- Copy is a real interaction, verified end-to-end: both popover variants
put the exact canonical npub on the actual clipboard — never the raw hex
the popover also lists, never a truncation — and a portaled popover's
clicks no longer steal focus from the new-DM To-field mid-copy. Pointer
copy, a natural Space-then-Enter path, and inner/outer Escape are
covered.
- Anything that isn't a valid identity key fails neutrally: short or
corrupt values — including degenerate values that technically encode to
a checksum-valid npub but aren't real identity keys — show "Unavailable"
with no copy button, instead of a misleading value.
- Both valid npub spellings display: all-lowercase `npub1…` and
all-uppercase `NPUB1…` (Bech32, npub's encoding, permits either casing)
both render the same canonical lowercase npub. Mixed case is rejected by
the display path as written — `canonicalNpub` and the widget don't
case-normalize input — while input parsing (`parsePubkeyInput`) keeps
its trim-and-lowercase normalization and accepts mixed-case npubs; both
paths require the decoded payload to be exactly a 64-character identity
key.
- Identity-key input is strict on payload: an npub whose decoded payload
isn't exactly a 64-character identity key is rejected, matching the
validation the app's Rust side already applies to agent allowlists.

Intentional scope boundary: only surfaces that render through the shared
widget change here. Outer profile copy, settings identity cards, the
respond-to allowlist, and workflow key fields still show hex — they move
to npub in the controls follow-up (#7489). Nothing else changes identity
representation: display names, private keys, event IDs, and the hex the
app stores, sends, and matches internally are untouched; only the
user-facing spelling of an identity key changes.

## Details

- `desktop/src/shared/lib/pubkey.ts` — `canonicalNpub()`: strict
canonical full-npub helper (64-char hex in any case, or a
checksum-validated npub, returns the canonical npub; anything else
returns `null`); `truncateNpub()`: the compact display form; existing
exports unchanged.
- `desktop/src/shared/ui/PubKey.tsx` — the shared widget's identity gate
validates through `canonicalNpub`; the popover copies the npub only.
- `desktop/src/shared/lib/nostrUtils.ts` — `parsePubkeyInput` rejects
npubs whose payload is not exactly a 64-character identity key.
- `desktop/src/features/messages/ui/NewMessageScreen.tsx` — the To-field
focuses its search input only for clicks that land inside the field
itself, so portaled recipient popovers keep their focus while open (a
popover click previously dismissed it mid-copy).
- Unit suites cover the helper, widget, and parser (including the
degenerate-encode and uppercase regressions); the e2e specs that render
these rows assert the npub display.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for npub identity display —
no existing match.
- Stack: #7489 is based on this branch and builds on these primitives;
it does not stand alone on main.

### Testing

At head `b3310c248` (base: main `44316ff72`; 12 files, +440/−39):

- Focused unit suites (pubkey, PubKey, parsePubkeyInput): 20/20 green;
mutation-checked — removing the decoded-length predicate fails the
short/empty checksum-valid-npub assertions in `canonicalNpub` and the
widget, and a wrong-identity clipboard value fails the new copy
assertions.
- `pnpm typecheck` and `pnpm check`: pass; full desktop unit suite
6459/6459 at this exact head.
- Targeted e2e at this exact head: 8/8 across the two specs that own the
clipboard flows — `agent-access-warning.spec.ts` (compact variant,
agent-access owner hint) and `pubkey-display-screenshots.spec.ts` (full
variant, new-DM recipient verification: pointer copy, popover surviving
the copy, inner/outer Escape, Space-then-Enter).
- No Rust-side or build files change in this PR, so those results are
unaffected.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): npub identity displays for mention, member, and workflow surfaces (#7495)

🤖
## Summary

Every Buzz account is identified by a long public key. Before this
change, when someone had no display name, surfaces fell back to
inconsistent labels — mostly raw hex fragments like `abcd1234…wxyz`,
sometimes a generic role label with no key — so the same person looked
different from surface to surface, and nothing looked like an npub
address. This PR applies the npub identity foundation from #7488 to the
everyday surfaces: a person without a display name now falls back to the
same compact npub everywhere — `npub1xxxx…yyyy`, the human-readable
spelling of their public key (first 8 + last 4 characters of the full
npub) — across messages and mentions, reactions, huddles, member and
participant lists, the sidebar and channel activity, search, projects,
tray, notifications, and workflow surfaces.

- **Mentions and messages**: key-only mention chips render the compact
npub. Pasting a copied mention back still re-binds it byte-exactly to
the identity it declares, for both the new npub chips and legacy
hex-truncated chips copied by older clients — wrong, missing, or
tampered key qualification is rejected instead of silently degrading to
plain text.
- **Reactions and huddles**: huddle reaction events and the huddle
roster/participants render the compact npub for unnamed participants;
workflow reaction triggers describe authors with the same form.
- **Members and sidebar**: channel and community member lists,
add-member results and invites, the members sidebar, the
channel-activity popover, search, projects (assignees/reviewers/PR
panels), the tray menu, and desktop notifications all fall back to the
compact npub; titles and aria labels keep the machine-readable full
labels.
- **Profile labels**: panel/popover display names and owner handles fall
back to the compact npub (never raw hex) when there is no name;
linked-event (nevent) message metadata shows the npub-shaped author
fallback while the event lookup and event IDs are unchanged.
- **Workflows**: author-picker secondary labels, step destination keys,
and trigger-author references render compact npubs; event and blob IDs
keep their existing hex compacts (they are not identities).
- **Avatars stay distinct**: fallback avatars for key-only identities
derive initials from the key's tail, so prefixed role labels like
"Participant npub1…" no longer collapse every unnamed participant onto
the same initials; people with names keep their name initials.

Preserved exactly: display names and distinct avatars, internal hex keys
(storage/API forms unchanged), clipboard identity roundtrips, event/blob
ID compaction, private keys (no nsec path is touched), and nevent link
handling.

Scope: this PR changes what identity labels **display**, not identity
controls — profile/settings copy controls, the respond-to allowlist,
workflow key fields, and agent dialogs are the sibling slice #7489, and
the shared primitives (`canonicalNpub`, `truncateNpub`, the `<PubKey>`
gate, strict input parsing) come from the foundation #7488.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for duplicates — none found;
the related work is the npub identity stack this slice belongs to.
- Base/dependency: stacks on #7488 (foundation) — this PR does not stand
alone on main.
- #7489 is a sibling slice on the same #7488 base
(profile/agent/workflow controls), not a dependency: this PR does not
require #7489, and #7489 does not require this PR — both only require
#7488.

### Testing

At exact head `4763cbeae1dd521309755e6d61f657324cb98667` (base:
`fix/desktop-npub-identity-d1a` @
`5f3a4a8111998c8aa41ad77cf66992bd1c85343c`; 71 files, +656/−189 —
production +277/−136, test support +379/−53):

- At this head: targeted `mentions.spec.ts` (1/1), the e2e build,
typecheck, and biome — green.
- 9 changed/related unit files: 100/100 green; typecheck, e2e build,
biome, and px text/truncation checks clean; huddle-roster focused run
green; channel-activity e2e 11/11; mutation checks confirm the fallback
wiring (removing it collapses shared initials and drops fallback rows).
- Known pre-existing local e2e failures, unchanged by this PR and
reproduced identically at the upstream merge-base: huddle-transcription
voice-menu attribution (25 pass / 1 fail) and the
`workflow-local-controls` 438px caret drift. Not claimed green locally.
- Update at head `236af9e6137386737e84d3a474d6bc808a704c50` (test-only
follow-ups `1143af345` + `236af9e6`): the `workflow-local-controls`
races were fixed in the test drivers, and the 438px diff was shown to be
a stale Darwin snapshot baseline (name-row enable switch already absent
and `message_posted` already MessageSquare at recording commit
`9390e11c9`) and refreshed — the focused screenshot test, including
keyboard/caret assertions, now passes locally (twice). The full spec was
not rerun after the snapshot refresh; the huddle-transcription item
above is unchanged.

Label/copy text changes are asserted by the e2e specs (`mentions`,
`mention-recipients`, `pubkey-display-screenshots`,
`huddle-transcription`, `channel-activity-popover`,
`workflow-local-controls`) rather than new screenshots; the screenshot
spec pins the compact npub text forms.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): npub identity controls across profile, agents, and workflows (#7489)

🤖
## Summary

Building on #7488's npub foundation, this PR finishes the identity
display change for the controls where you actually manage people and
keys: profile, settings, agent access, and workflows. Everywhere in
these surfaces, an identity key shows — and copies — as its canonical
npub (npub is the human-readable encoding of a Nostr public key: the
compact `npub1j57...fjmv` form where space is tight, the full npub where
the whole key matters), and accepts npub as input.

After this change:

- Profile panel: the public-key row and the managed-by / declared-owner
copies show the full npub. If a key can't be encoded, you see
"Unavailable" with no copy button — never a raw or partial key.
- Settings: the identity card shows and copies the npub. The
hosted-communities account identity derives from the bound key
(`pubkey_hex`) — the same authority as the mismatch gate and hosted
operations — so the display can never disagree with what the app acts
on; an unusable hex falls back to a neutral label instead of rendering
the unverified server npub. The connected claim and a community's
Connect action require that same usable bound key to match the local one
— with no usable binding the card cannot claim connected or start
Connect, while the community list, linking, and delete/rebind recovery
stay available.
- Hosted create/onboarding: the account and device identity rows in the
create flow and owner onboarding derive from the same authoritative
fields (bound key / local key), with the same neutral fallback;
readiness requires a usable bound key that matches the local one.
- Respond-to allowlist (controls who may respond to an agent): entries
can be typed or pasted as hex or npub; both spellings of the same key
are recognized as one entry and dedupe. Search results, chips, and
remove buttons use the compact npub.
- Workflow key fields: to/from keys display as npubs in the form and
save back as canonical hex. Templates like `{{trigger.author}}`, roles,
and free text pass through untouched; placeholders accept both
spellings.
- Recipient and agent dialogs: the verify popover is npub-only (the
raw-hex line is gone); denied-membership screens never show a raw key.
- The Rust-side truncated display name (used for native surfaces) shows
the same compact npub, so those surfaces match the web UI.

Internal representation is unchanged: keys are still stored, sent, and
matched as canonical 64-character hex — npub is a display and input
spelling, normalized to hex at the boundary, so existing data and
integrations keep working. Bound-key usability and comparison use one
normalized form (trimmed, lowercased, 64 hex characters; npub rejected),
so padded or mixed-case spellings of the same key match. Display names,
private keys, and event IDs are untouched.

## Details

- `respondToAllowlist` / `RespondToField`: npub entries normalize to
canonical hex; cross-form dedupe; compact npub in rows and chips;
direct-add accepts npub and stores canonical hex.
- `workflowFormTypes` / `WorkflowStepCard`: hex → npub for display, npub
→ canonical hex on save; templates, roles, and free text pass through in
both directions (roundtrip-tested).
- `UserProfilePanelFields`, `ProfileSettingsCard`,
`HostedCommunitiesSettingsCard`, `MembershipDenied`,
`SelectedRecipientChip`, `AddAgentToChannelDialog`: npub display and
copy; invalid keys → "Unavailable" with no copy; hosted identity rows
derive from the bound `pubkey_hex` (create/onboarding rows from the
bound and local keys), never the unverified server npub;
connected/readiness/Connect gates use the same usable-bound-key
predicate, and the settings Connect invocation callback re-checks it
before starting.
- `src-tauri/src/commands/identity.rs`: `truncated_display_name`
compacts to the first 8 + last 4 characters of the npub (above a 12-char
threshold), mirroring `truncateNpub`.
- e2e: profile key rows and clipboard polls assert npub forms and
raw-hex suppression; the display-screenshots spec pins the npub-only
popover; hosted specs drive the real settings card, create flow, and
onboarding rows through their real providers, and the unlinked/npub-only
identity cases assert no connected claim and no Connect action.

### Related issue

- Fixes: N/A. No separate issue; the related work is the stack below.
- Stack: builds on #7488 (shared npub foundation), now merged; this PR
is rebased onto main and stands on its own.

### Testing

At head `303c90ffa` (base: main `bfc38485`; 24 files, +1125/−146):

- Focused unit suites (respondToAllowlist, workflowFormTypes,
hostedCommunityApi bound-key helpers) green; mutation-checked — dropping
allowlist canonicalization fails the dedupe case, and dropping bound-key
normalization fails the npub-in-hex and padded same-key cases.
- Full desktop unit suite 6,477/6,477, `desktop-typecheck`,
`desktop-check` (formatting fixed narrowly with `biome check --write` on
the touched files only), and a fresh E2E build at the current head; the
add-community + hosted-communities-settings specs 18/18 and onboarding
integration 69/69 on a fresh dedicated port, with focused new-case runs
4+4 covering padded same-key (ready, Connect kept — no false rebind) and
npub-in-hex (neutral label, recovery, no Connect) across the settings
card, create flow, and first-community onboarding, plus the
unlinked-account settings regression asserting Connect cannot occur.
- `cargo fmt`/clippy (both feature sets) and `cargo test identity` (71
pass) passed at the earlier full-change head; since then, the only
production changes in this PR's delta are the hosted identity display
authority and its fail-closed bound-key gating/normalization above
(base-side fixes carry #7488's receipts) — every other change is
test-only.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(mobile): standardize public-key identity display on npub (#7493)

🤖

## Summary

In the mobile app, anyone who hasn't set a display name shows up as a
raw 64-character hex key (e.g. `3a5d4f9c…`) — unreadable, and
unrecognizable as the same identity across screens. Profile and Settings
also let you copy that raw hex. Nostr public keys have a standard
readable form — `npub1…`, the same encoding other Nostr apps and our
desktop app already display. This PR makes every mobile identity surface
render npub instead:

- **Unnamed people everywhere** — message and thread authors, reactions,
typing indicators, member lists, channel details, DM headers and tiles,
inbox, search, forum cards, Pulse notes and reply context, mention
suggestions, and invite rows — now show a compact npub label: first 8 +
last 4 characters of the full npub joined by an ellipsis
(`npub1abcd…wxyz`), the same truncation desktop uses. Previously these
showed truncated raw hex.
- **DM fallback avatars and blank names** — 1:1 DM tiles and headers key
their fallback avatar to the same non-self counterpart the label names,
including self-first participant order; a self-DM keeps its
hex-key-derived initial. Blank or whitespace-only display names fall
back to the compact npub instead of rendering empty, while nonblank
authored names render verbatim (padding included).
- **Profile sheet → "Copy public key"** now copies the full canonical
npub — never raw hex. When the identity string isn't a valid public key,
the copy tile is disabled, so a malformed key never reaches the
clipboard.
- **Settings → Identity (pubkey)** displays and copies the full npub; an
invalid identity reads "Identity unavailable" with copy disabled.
- **Invalid identities never leak truncated raw hex** into the UI
anywhere — they render a neutral "Unknown identity" label.
- **Unchanged on purpose:** display names and verified handles (NIP-05 —
the `name@domain` badge) still render as before. Unnamed avatars keep
distinct per-key initials, derived from the underlying hex key rather
than the npub — otherwise every unnamed key would render the same "N"
initial. Event IDs are not public keys, so they keep their hex
truncation (in Pulse's "Replying to", the parent author shows npub while
an event-id fallback still shows hex). The nevent share link, private
keys, and internal hex storage are untouched. Inputs that accept a key
(invite/member entry) accept both hex and npub and keep working in hex
internally.

### Related issue

N/A. Searched open issues/PRs for npub identity display on mobile —
closest related: none found. Desktop's parallel npub standardization
lives in the stacked desktop PRs (#7488 foundation, #7489 controls,
#7495 display surfaces); this is the independent mobile slice (based
directly on `main`, not on those branches).

### Testing

At exact head `5a620e420a1fd57d9d8011ac26434eed32fcf765` (base: `main`
`44316ff72`; 40 files, +1,345/−154):

- Full mobile suite: 2,098 tests passing (`cd mobile && flutter test`);
`flutter analyze` clean; `dart format --set-exit-if-changed .` clean —
the same checks CI runs.
- Widget/unit coverage at production seams: compact labels and hex-keyed
avatar initials for DM headers/tiles, member rows, mention suggestions,
and Pulse reply context; DM fallback avatars keyed to the labeled
counterpart (self-first order and self-DMs); blank/whitespace
display-name npub fallback with nonblank authored labels verbatim,
including the Activity inbox sender and profile-sheet heading (each with
its own empty/whitespace production-seam regression); full-npub copy and
disabled-copy semantics in profile and settings; invalid-key
suppression; and hex↔npub input round-trips.

Verified via unit and widget tests — no device/simulator validation is
claimed.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): order unnamed roster members by full canonical npub (#7503)

🤖

## Summary

- Channel members appear in the Members sidebar. A member who has never
set a display name is listed under an abbreviated form of their public
key (npub), and the sidebar previously sorted those unnamed members by
that short label. Short labels are not unique — different keys can share
one — so the order of unnamed members could look arbitrary or unstable.
Unnamed members now sort by their full public key, so the order is
deterministic.
- When two members display the same name, the previous tiebreak was
membership order (who joined first), which is not visible to a reader
and can shift as roster data loads in. The tiebreak is now the full
public key, so identical display names always land in the same order.
- Nothing gets noisier on screen: the full key is used only for sorting,
and the sidebar still shows the compact abbreviated form. Priorities are
unchanged — authored (custom) names still outrank fallback labels, and
role/current-user grouping still applies.
- Scope is the desktop app's Members sidebar and member management: the
two existing sort comparators. Mobile and other lists in the app are
untouched.

### Related issue

Based on #7495 (introduced the abbreviated npub labels this follows up
on). The original five presentation PRs remain independently reviewable.
No closer duplicate found.

### Testing

- 6469 desktop unit tests, typecheck, and check pass.
- The 3 existing consumer-seam E2E tests still pass; a new E2E test
asserts the sidebar lists unnamed members in full-key order, with
fixture members deliberately inserted in the opposite order so incoming
membership order cannot mask the sort.
- Negative check: reverting only this change makes the new ordering
assertion fail, so it genuinely binds the new sort.
- CI has not run on this PR yet.

Buzz provenance: channel 1f0e4a3d-7e01-4efe-bb16-843b357f85c9 / task
340c3de9b27dbedb8453c0c7652220f9080d30fcc70a7c4f6e27fdd4fa378056

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): require a Codex adapter with Astra support (#7427)

## Summary

Buzz considers codex-acp 1.6.2 current because the supported adapter
floor is still 1.1.7. That adapter bundles Codex 0.148.0, so updating a
separate Codex CLI to 0.153.4 leaves managed agents on the older runtime
and unable to use GPT-6 Astra.

Raise the supported adapter floor to the published 1.10.0 release, which
depends on `@openai/codex ^0.153.3`. Existing discovery and installation
code then classifies older adapters as outdated and offers the managed
reinstall path. Update the availability and install-plan regressions to
cover the observed 1.6.2 installation and the new minimum.

This follows the existing version-floor policy. It does not
automatically update a running installation: the user must complete
Buzz’s offered adapter upgrade. Future upstream compatibility changes
may require another floor update.

### Related issue

No exact duplicate found in searches for Astra, CODEX_PATH, bundled
Codex, outdated runtime, and codex-acp 1.10. Related: #3097 raised the
older floor to 1.1.7 (already present on main); #2422 covers lost error
details for runtime mismatches. Neither resolves this version gap.

Originating conversation:
buzz://message?channel=3286cd76-f83e-4c7d-8317-10a16580744d&id=8b79a73078217222b870fff144c27e7d27bcd5a67c966869c18fe726db716898

### Testing

- Isolated npm install of codex-acp 1.10.0 resolved bundled Codex
0.153.4, with no CODEX_PATH override.
- Live macOS ACP probe: initialize protocol v1 → session/new → select
gpt-6-astra[medium] → prompt. Received `OK` and `stopReason: end_turn`;
usage metadata confirms gpt-6-astra.
- Existing adapter 1.6.2 initialized but advertised no Astra model in
the same probe.
- Desktop Rust formatting and `git diff --check` pass.
- `just desktop-tauri-test`: 3,266 passed, 20 ignored, zero failures
across the Desktop workspace and integration tests.
- Workspace and Desktop Clippy, frontend static checks, and `just
file-size-check` pass.
- Repository `just ci`: still running the remaining
mobile/build/workspace-test stages.

The installed Buzz app and managed adapter were not replaced or
restarted. The live check validates the new adapter/runtime path; a
complete packaged Desktop upgrade workflow remains untested.

Signed-off-by: Stephen DeLorme <[email protected]>

* fix(buzz-acp): report missing models without retrying (#7538)

## Summary

When an agent reports model-not-found, Buzz retries the unavailable
model and delays the failure reply until retries are exhausted. Stop
retrying this error and immediately post a threaded recovery notice. The
notice tells users to select a different model in agent settings, save,
restart the agent to apply the configuration, and re-send their request.

This adds one error-handling branch and regression coverage in
`buzz-acp`. It matches `-32002` errors containing `model not found`.
Other resource-not-found errors, such as stale sessions, retain the
existing retry behavior. Detailed error events remain available for
diagnosis. The existing restart policy is unchanged.

### Related issue

None found in existing issue/PR searches for model-not-found recovery.

### Testing

Playwright captured and visually checked the thread UI with seeded
conversation data and the exact recovery text. The check opens the
request's thread, confirms no reply before the failure, injects the
notice, and verifies the full text is visible. [Before/after
screenshots](https://github.com/block/buzz/pull/7538#issuecomment-5608196506)
show the corrected save-and-restart instructions. These are local test
captures, not a deployed provider recovery flow.

Generated with Codex

---------

Signed-off-by: Diem Nguyen <[email protected]>

* fix(desktop): let inbox title and message author names truncate under narrow panes (#7550)

## Summary

Fixes two instances of the same dead-truncate pattern in the desktop
app, where a flex item's implicit `min-width: auto` prevented `truncate`
from engaging, so long text painted over adjacent controls instead of
ellipsizing:

- **Inbox detail title** (`InboxDetailPane.tsx`): the clickable
context-title button sized to its text instead of shrinking with the
pane, overlapping the header controls (open-in-channel, members, huddle,
more menu). Fixed by adding `max-w-full`.
- **Message author names** (`MessageHeader.tsx` /
`UserProfilePopover.tsx`): the `UserProfilePopover` inline-flex trigger
wrapper refused to shrink below the name's nowrap width, running long
author names under the hover action bar and off the pane edge. Fixed by
adding a `triggerClassName` prop to `UserProfilePopover` and passing
`min-w-0 max-w-full` at the author call site.

Two other suspected instances (project file breadcrumb, drafts pane
title) were stress-tested and already truncate correctly — no change.

### Related issue

N/A — none found.

### Testing

- New Playwright regression tests for both fixes
(`inbox-title-overlap.spec.ts`, `message-author-overlap.spec.ts`,
registered in the smoke project), each proven to discriminate: they fail
with the fix reverted (real measured overlap) and assert the ellipsis
actually engages with non-zero title width, so they can't pass
vacuously.
- Typecheck, lint, and full desktop unit suite green (pre-push hooks);
full desktop e2e smoke suite run earlier: 1402 passed, 3 pre-existing
unrelated failures (each fails identically with the fix reverted).

**Inbox title — before** (long title paints under the header controls):

![Inbox title before: title text overlaps the header control
icons](https://github.com/user-attachments/assets/d5a10f98-114f-4466-8012-468616b82807)

**Inbox title — after** (truncates with ellipsis, controls stay clear):

![Inbox title after: title truncates with an ellipsis before the
controls](https://github.com/user-attachments/assets/80520f7c-1120-4dbc-929d-1ef5eceb874b)

**Author name — before** (long name runs past the header row edge):

![Author name before: name glyphs bleed past the action
bar](https://github.com/user-attachments/assets/be9743ae-ef51-4db0-8c46-a0649e98f0d5)

**Author name — after** (clean cutoff):

![Author name after: name truncates cleanly inside the header
row](https://github.com/user-attachments/assets/24cc0412-ca99-40c4-add6-13380b3ae065)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Signed-off-by: cynfria <[email protected]>
Signed-off-by: Tree Trunks <6ba22921d9dc2ad0aa6ecdf63787ddd24726e266d866da31af69f2e4e146ace5@buzz.block.builderlab.xyz>
Co-authored-by: Claude Fable 5 <[email protected]>
Co-authored-by: Tree Trunks <6ba22921d9dc2ad0aa6ecdf63787ddd24726e266d866da31af69f2e4e146ace5@buzz.block.builderlab.xyz>

* fix(relay): reject presence updates when Redis storage fails (#7532)

## Summary
- Reject kind:20001 presence events with `OK false` / `error: presence
storage unavailable` when Redis SET or DEL fails, before publishing,
local fan-out, or local-event marking.
- Preserve the producer contract needed by snapshot-confirming
consumers: delivered live presence must follow successful mutation of
the Redis state read by snapshots.
- Classify those backend rejections with the existing `IngestError`
taxonomy so a presence storage outage counts as
`buzz_events_rejected_total{transport="ws",reason="error"}`, not client
`reason="invalid"`; genuine client-input refusals (verification failure,
membership gates) stay `invalid`, and every wire message is an unchanged
fixed sanitized string (review follow-up, no protocol wording change).
- Add actual `handle_event` integration coverage for rejected
online/offline transitions, healthy online→offline
accepted/stored/fanned-out behavior, and the rejection-counter routing
on storage failure with an invalid-signature control.

This is standalone on main; it does not depend on the mobile
implementation. Deploy this relay prerequisite before relying on #7526's
snapshot-confirmation policy. Existing
pubsub-failure-after-successful-storage behavior and disconnect TTL
cleanup are deliberately unchanged. A storage error may be an ambiguous
write outcome, not a rollback guarantee; the rejected event is not
published by this handler. Clients may retry the generic `error:`
rejection. Desktop's 60s heartbeat retries non-offline presence, not
every explicit offline transition.

### Related issue
Addresses the relay prerequisite identified in [#7526 review
5157607827](https://github.com/block/buzz/pull/7526#pullrequestreview-5157607827).
Searched open presence/storage PRs; no duplicate relay storage-error
rejection fix found. #7382/#7383/#7526 heads and bases are unchanged.

### Testing
Exact head: `389174df29cc02d0f885c03209eff661d8bb2ec0` (+380/-13; 393
total), one commit `389174df2` on top of the reviewed `c031d6eb1`
(DCO-signed; base `bfc384855889432df4a333a0edf3080f332ee169` unchanged).

- PASS: `cargo fmt --all -- --check`, `cargo clippy -p buzz-relay
--all-targets -- -D warnings`, `git diff --check`, `just
file-size-check`, PostgreSQL discovery validation — all run at the exact
final head with a clean tree before and after.
- PASS: documented native `scripts/postgres-test-run.sh -p buzz-relay
--lib --tests`: **89/89** actual integration tests, including the four
presence cases (online/offline storage rejection, healthy
online→offline, and the new rejection-classification case). Owned
PostgreSQL 17/Redis on isolated loopback ports, schema plus
reconciliation applied; no shared development database.
- PASS: explicit `cargo test -p buzz-relay presence_storage -- --ignored
--nocapture`: **4/4**, not skipped.
- Full isolated relay crate suite at the final head (`cargo nextest run
-p buzz-relay --lib --tests`): **1062 run: 1062 passed, 94 skipped**.
The previously failing
`api::mesh_demo::tests::demo_join_forwarded_arm_round_trips_echo` passed
in this run (1.5s); it is a known timing-sensitive main baseline failure
tracked open in #7140 and untouched by this PR, so this single passing
run is reported as-is and does not claim environmental clearance or
close #7140. No full-suite-green claim is made beyond this run.
- Mobile is untouched; #7526's existing 2090-test/format/analyze
evidence remains scoped to its unchanged head. Its separate Desktop
Smoke E2E (2) failure remains red; no CI retries requested.

[Production-seam regression
coverage](https://github.com/block/buzz/blob/389174df29cc02d0f885c03209eff661d8bb2ec0/crates/buzz-relay/src/handlers/event.rs#L1491-L1803):
the metric case drives real `handle_event` traffic against a genuinely
dead Redis endpoint with a seeded active PostgreSQL community and a
registered presence watcher, asserts the storage rejection counts
`reason="error"` while a tampered-signature control through the same
dispatcher arm stays `reason="invalid"`, and re-asserts the rejected
ACK, no fan-out, and no local-event marker. Counter assertions use a
thread-local recorder guard held across `.await` points (the buzz-db
counter-test convention) inside the per-process nextest postgres-ci
lane, so no parallel test can race the counter snapshot.

No UI change or screenshot. Local logs and reproducible service/gate
scripts are retained under
`WORK_LOGS/MOBILE_FEEDBACK_PRESENCE_20260909/relay_prerequisite/metric_correction/`
in the engineering workspace. This PR is a review candidate, not merge
clearance.

Causal checks: restoring only the pre-fix production mutation block
makes both original rejection tests fail (`OK true` instead of `false`);
healthy success still passes. Reverting only the typed classification
(mapping the ephemeral `Internal` arm back to `invalid`) makes the new
metric regression fail with the outage counted as `[("ws","invalid",2)]`
instead of `[("ws","error",1),("ws","invalid",1)]`. The unchanged mesh
echo case also failed 504/200 with the main-production block restored in
the prior run, supporting its separation from this change without
claiming environmental clearance. Candidate source restored
byte-for-byte after each mutation. Repository-wide `just ci` was not
rerun; the scoped relay gates above are the new evidence.

---------

Signed-off-by: Logan Johnson <[email protected]>

* fix(markdown): align mention chip wrapping (#7501)

**Category:** fix
**User Impact:** Human and agent mentions now break across lines with
the same cloned chip treatment as repository and permalink chips while
preserving the conversation text rhythm.

**Problem:** Profile-backed rendered mentions sat inside an
`inline-flex` popover trigger, unlike entity chips, so the wrapper
interfered with true inline fragmentation. The browser-layout test
measured text-range rows rather than the painted chip rectangles,
allowing touching decorations to pass as “separate” fragments.

**Solution:** Keep the profile trigger interactive but override its
layout to true `inline`, then give mention fragments 18px computed
leading inside the message’s 20px prose rhythm. Chromium paints each
fragment at 17px and advances it by 20px, leaving a visible gap between
cloned rounded rectangles. The browser test now measures the chip’s own
`getClientRects()` and asserts fragment count, height, gap, and step;
entity links retain their existing 22px leading.

<details>
<summary>File changes</summary>

**desktop/src/features/profile/ui/UserProfilePopover.tsx**
Allows inline consumers to override the trigger wrapper’s layout without
changing other profile-popover call sites.

**desktop/src/shared/styles/globals/markdown.css**
Keeps one shared wrapping-chip mechanic and gives mention decorations
enough room to separate visibly within 20px prose.

**desktop/src/shared/ui/markdown.test.mjs**
Pins both rendered mentions and entity links to the shared wrapping-chip
contract.

**desktop/src/shared/ui/markdown/MarkdownMention.tsx**
Makes the profile-popover trigger truly inline so the nested mention
chip can fragment with surrounding prose.

**desktop/src/shared/ui/mentionChip.ts**
Keeps `wrapping-inline-chip` as the single contract for fragmenting
decorated chips.

**desktop/tests/e2e/mentions.spec.ts**
Measures the painted chip rectangles, requires a positive fragment gap,
and verifies the inline trigger remains mouse- and keyboard-operable.

**desktop/tests/e2e/navigation.spec.ts**
Keeps a wrapped repository chip as the control, asserting its existing
22px line height and fragment advance.

</details>

## Reproduction steps

1. Open a channel in Buzz Desktop using dark theme.
2. Send a message containing a human mention and another containing an
agent mention; both chips should remain aligned with adjacent text on a
20px line.
3. Render a collision-qualified mention in a narrow message width; it
should break into separately decorated fragments exactly like another
wrapping chip, while each fragment follows the 20px prose rhythm.
4. Render a long repository or permalink chip in the same constrained
width; it should retain its roomier 22px fragment spacing.

## Screenshot

The dark-theme production renderer shows the real qualified label (`bob
(npub1hv3…tpuc)`) at an 8rem width. The two lines now paint as visibly
separate rounded fragments rather than one continuous rectangle.

![Qualified human mention wrapping into two visibly separate rounded
fragments in the dark-theme Buzz
timeline](https://github.com/user-attachments/assets/ee6e7be5-937e-4022-9c82-cf71f8203470)

## Validation

At commit `2b063e1b4ade30e11f1616269ad4ba4190366885`:

- Pre-push desktop gates — file-size check, Biome/checks, typecheck, and
6,483 unit tests passed
- `pnpm --dir desktop build` — passed
- Focused Playwright coverage for single-line agent mention, single-line
human mention, wrapped qualified mention including keyboard profile
activation, and timeline mention click — 4 passed
- `git diff --check` — passed

---------

Signed-off-by: Taylor Ho <[email protected]>
Co-authored-by: Rizz <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>

* fix(acp): integrate the Buzz Pi adapter fork (#7552)

## Summary

PR #7335 worked around missing Pi adapter support by generating a
private Pi launcher and injecting Buzz's standing prompt and skills at
process launch. The Buzz Pi fork now carries the required adapter
extensions, so this removes that launcher and returns prompt
construction to the normal ACP session path while retaining the
base-prompt composition introduced by #7335.

The Pi preset now installs `salman1993/pi-acp` and launches its renamed
`buzz-pi-acp` binary. Buzz adds `-- --skill
<harness-cwd>/.agents/skills` only when launching that binary, sends the
complete composed prompt as the `_meta.systemPrompt` replacement string
on `session/new` only when `initialize.agentInfo.name` is `buzz-pi-acp`,
and sends the scoped title alongside it as `_meta.sessionTitle`. The
fork identity is treated as system-prompt capable regardless of its
reported ACP protocol version, which prevents duplicate legacy
user-message framing. Upstream `pi-acp` does not receive either
fork-specific behavior. Observer transcript projection accepts the
string, `{ replace }`, and `{ append }` metadata forms.

The fork now stores restore metadata in one atomic file per session
under `~/.pi/buzz-pi-acp/sessions/`. This prevents concurrent Buzz
workers from overwriting another session's prompt or title. The fix
landed in
[salman1993/pi-acp#9](https://github.com/salman1993/pi-acp/pull/9).

This supersedes the closed #7508. No agent-configuration rules changed;
this changes the Buzz Pi adapter contract and launch arguments.

### Related issue

#7329

### Testing

Installed fork commit `09cf07e436b8f18e52401558f988f31a15702313` through
the documented Git URL. The installed bundle matched the committed
`dist/index.js` byte for byte and contained the `~/.pi/buzz-pi-acp`
metadata path. The fork's 106 non-skipped tests, typecheck, and lint
pass.

Ran the ignored real-Pi integration test through Buzz's production
session composer. The test exercised the renamed package,
`agentInfo.name`, and the new per-session metadata store. Base, persona,
team, core-memory, huddle, canvas, and skill markers each appeared once
after switching sessions and again after restarting the adapter, while
the other session and Pi's native default prompt were absent.

Added regression coverage proving `buzz-pi-acp` receives fork-specific
system-prompt metadata and managed skills while upstream `pi-acp` does
not. `just ci` passes.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

* feat(git): add default-branch management to relay and CLI (#7562)

Authored by Brain and opened on behalf of Wes (`wesbillman`).

## Summary

Add `buzz repos default-branch get…

* fix(mobile): style inline code with the app mono face (#6631)

## Summary

Inline code on mobile renders as **bold body text on a faint background
wash** — no monospace face, no chip, and it cannot wrap. #5257 diagnosed
this as a missing `highlightBuilder`.

That is no longer the right fix. `gpt_markdown` 1.2.0 deprecates
`highlightBuilder` (removal in 2.0.0), renders inline code as a real
chip, and adds `InlineCodeStyle` for restyling it. The package author
confirmed this on the issue. So this PR is an upgrade — 1.1.6 → 1.2.1 —
plus one theme declaration, rather than the builder the issue originally
asked for.

**Where the style is declared.** `GptMarkdownThemeData` goes in
`AppTheme._buildTheme`, which both `light()` and `dark()` call. That
reaches all four `GptMarkdown` call sites — `message_content`,
`transcript_item_widget`, `token_pill`, `custom_emoji_render` — so the
style is stated once instead of per widget. A widget-level
`inlineCodeStyle` would have covered channel messages only, leaving the
other three on the package's defaults.

**What is declared.** Face, size, ink, chip fill and outline — not the
face alone. A face name on its own leaves the rest on the package's
defaults, which put inline code at 14.1sp beside a fenced block's 13, on
a neutral `onSurface` tint rather than the app's code surface. In dark
that tint is *lighter* than the surface, while every other code surface
in the app is recessed, so the chip read as a different kind of object.
All of it now comes from one `CodeStyle` declaration that the fenced
block reads from too, so the two cannot be edited apart.

**Three adaptations the upgrade requires.** Each was found by running
the gate, not by reading the changelog:

1. **`imageBuilder` widened** to `(context, url, width, height)`. This
is a hard compile error, and it is **not listed in the package's
migration guide**, which states "nothing here stops code compiling".
Worth reporting upstream.
2. **`autolink` now defaults to `true`.** `normalizeBareLinks()` already
rewrites bare URLs into Markdown links before rendering, so both would
run. `message_content` opts out with `autolink: false` to keep current
behaviour exactly. The migration guide argues for dropping the
pre-processor instead — a better fix, but a behavioural change that
belongs in its own PR.
3. **`gpt_markdown.dart` now re-exports `markdown_config.dart`**, making
two direct imports redundant. `flutter analyze` reports `No issues
found!` on 1.1.6 and flags both on 1.2.1, so these warnings are new, not
pre-existing.

**Deliberately out of scope.** The three non-message call sites now
autolink bare URLs, since only `message_content` has a pre-processor to
collide with. Custom inline components (`_MentionMd`, `CustomEmojiMd`,
`_ChannelLinkMd`) could additionally declare `allScopesExceptLinkLabel`
— 1.2.0 offers it as the fix for a `WidgetSpan` chip going blank inside
a link label on iOS — but current behaviour is unchanged without it, so
that stays a separate change.

### Related issue

Fixes #5257

Duplicate scan: searched `gpt_markdown`, `inline code mobile`,
`highlightBuilder` and `InlineCodeStyle` across both PRs and issues. No
open PR touches inline code styling. #6135 (link labels) and #6166 (text
selection) also touch mobile Markdown but address different defects.

### Testing

Full gate, `just ci` — exit 0:

| Stage | Result |
|---|---|
| Rust (33 suites) | 4768 passed, 0 failed |
| Desktop | 5799 passed, 0 failed |
| Mobile | **2011 passed**, 0 failed |
| `flutter analyze` | `No issues found!` |
| Desktop + web build | ok |

Run on the branch with `main` merged in, so these numbers match what CI
builds.

**New regression test** — `renders inline code in the app code style`.
It resolves the `CodeTextSpan` the package tags inline code with, which
carries both the resolved `TextStyle` and the colours the chip behind it
is painted with, so face, size, ink, fill and outline are all asserted
rather than a widget's presence. It is negative-controlled: reverting
only the theme declaration fails it with

```text
Expected: a numeric value within <0.001> of <13.0>
  Actual: <14.1>
```

and dropping the declaration entirely falls back to
`packages/gpt_markdown/JetBrainsMono` — so the test measures the real
thing, and it would catch a future regression that silently drops the
theme extension.

The test passes `baseStyle: messageBodyTextStyle`, the style the message
surfaces actually use; the widget's own fallback is the smaller
`bodyMedium`, which would move the expected size.

The test finds paragraphs with `find.byWidgetPredicate((widget) =>
widget is RichText)`, not `find.byType(RichText)`: inline code renders
through `BidiRichText`, a `RichText` subclass, and `byType` matches
exact runtime types.

That is a hazard for any test that reads text back out of a paragraph,
and one landed after this branch was cut:
`message_content_custom_emoji_test.dart` arrived with #6996 and its
`code keeps literal emoji while adjacent known tokens render` case reads
a code span through `find.byType(RichText)`. It passes on `main` and
fails on the merge result, which is what CI builds, so it went red only
once CI was authorized. It now uses the same predicate. The two other
`byType(RichText)` call sites — the rest of that file and
`message_author_meta_test.dart` — were re-run and pass: their content
carries no code span, so the exact type still matches. They were left
alone.

### Screenshots

Rendered through the real `MessageContent` widget with the app's own
fonts loaded, at 390pt wide, 3x DPR. Sample text: ``Set `BUZZ_RELAY_URL`
before launch, then run `just mobile-test` to verify.``

| | Before (1.1.6) | After (1.2.1) |
|---|---|---|
| Light |
![before-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-light.png)
|
![after-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-light.png)
|
| Dark |
![before-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-dark.png)
|
![after-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-dark.png)
|

Before: bold Inter on a flat wash, no chip edge, and `just mobile-test`
breaks across the line with the wash simply ending. After: Geist Mono in
a bordered, rounded chip, and the wrapped fragment gets its own chip on
each line.

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>

* fix(buzz-acp): wake held ACP threads and fence forked sessions (#7340)

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[#7337](https://github.com/block/buzz/pull/7337#pullrequestreview-5116329341).

### Related issue

Follow-up to #7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

* fix(agent): route GPT-5+ model-service FQNs to Responses (#7358)

## Summary
Route Databricks Unity Catalog model services to OpenAI Responses when
their service name matches GPT-5 or newer. These models can reject tools
plus reasoning on Chat Completions.

Match only the service component, using the existing family-token
boundaries and a numeric major version. Catalog and schema names cannot
select the protocol. Keep neutral effort capabilities and the full model
ID unchanged; other services still use MLflow Chat Completions.

Keep the Rust and desktop resolvers in sync, add shared boundary cases
and a captured-HTTP regression for completion and summarization, and
update the documented FQN rule.

### Related issue
No duplicate found in searches for “FQN responses” PRs or “astra”
issues. Related: #6918 introduced Unity Catalog discovery.

Originating conversation:
buzz://message?channel=0b881928-a3a6-4c01-b981-8e64268f01ce&id=770949343bc96a9ed88acd90a1b37d358a0efc52c79237d0fdb491ce02b8d4ed

### Testing
No live Databricks inference test. The gateway must accept the full
model-service ID on its OpenAI Responses route; this remains the
integration risk.

The local `just ci` attempt exceeded its five-minute deadline during
`mobile-check`, so the full repository gate was not completed. All
push-hook checks passed.

Generated with Codex

Signed-off-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>
Co-authored-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>

* feat(mesh): upgrade to mesh-llm 0.76.0-rc8 and recommend Qwen3.8 27B (#6189)

Upgrades Buzz's mesh-llm dependencies through the released `v0.76.0-rc8`
tag (`2040765d`), including the Qwen3.8 curated recommendation and rc8's
scheduler/runtime improvements.

**Scope note:** the earlier open-relay/unenforced-admission mode has
been removed from this PR at Mic's direction — it is not a product mode
we want. Mesh admission remains roster/allowlist driven, exactly as on
`main`: on a relay with no NIP-43 membership snapshot the mesh runs
self-only. No NIP-11 mode probing, no mode-transition restarts. A future
perimeter/admission strategy for open relays will be designed
separately.

This PR also:
- seeds `BUZZ_AGENT_LLM_TIMEOUT_SECS=660` for mesh agents, above
MeshLLM's 600-second backend timeout;
- makes `desktop-tauri-clippy` lint both default and `mesh-llm` cfg
graphs;
- runs the feature-enabled desktop test suite in CI;
- recommends Qwen3.8 27B Q4_K_M for 64 GB-and-larger machines, then
ladders down through Gemma 4 E4B and Qwen 9B for smaller machines;
- keeps stored shared-compute `auto` translated to MeshLLM's supported
wire model `mesh`.

RC8 verification:
- `just ci` passed locally at
`92ecc7ec933bdd4df804cc9f28a2b51efa5313c5`.
- Pre-push differential gates passed, including both desktop Tauri cfg
graphs and package tests.
- A prior isolated runtime smoke used the RC8 binary's OpenAI endpoint
for a Buzz-shaped system/user/tool/tool-result/final-response loop; all
assertions passed and the isolated process was shut down.

Perf previously measured on M5 Metal, Qwen3.8-27B-Q4_K_M: TTFT 0.22–0.32
s, ~25 tok/s streaming; agent-shaped turns ~1 s to first token after the
first (prefix cache).

---------

Signed-off-by: Michael Neale <[email protected]>
Signed-off-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Signed-off-by: Alessandro Joabar <[email protected]>
Co-authored-by: Michael Neale <[email protected]>
Co-authored-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Co-authored-by: Mic Neale <[email protected]>
Co-authored-by: Alessandro Joabar <[email protected]>

* fix(link-preview): keep composer fetches user-paced (#7211)

**Category:** fix
**User Impact:** Link previews can keep loading while a message is being
composed, while sending still has a finite escape hatch and stalled
network transports cannot occupy preview slots forever.

**Problem:** Native metadata and image deadlines could collapse slow
previews into fallback cards while the user was still composing, and a
shared image-host cooldown made pasted batches fail inconsistently after
one rate limit. **Solution:** Keep preview resolution user-paced with no
aggregate request deadline, bound transport inactivity (15s DNS/connect,
30s idle read), serialize image requests by host, and allow at most one
server-directed cooldown wait of up to 30s across an image fetch and its
redirects. The existing bounded post-Send preparation and immediate Skip
paths remain unchanged.

<details>
<summary>File changes</summary>

**desktop/src-tauri/src/commands/link_preview.rs**
Removes aggregate native deadlines so composer metadata work can
complete at the user's pace, while retaining DNS/connect/idle-read
liveness bounds. Adds bounded host-paced image request coordination that
releases its gate during cooldown, waits inline at most once for at most
30 seconds, and cannot renew that wait through redirects or the outer
transient retry. Same-host image and favicon requests remain
deliberately serialized to align with host rate limits.

**desktop/src-tauri/src/commands/link_preview_rate_limit.rs**
Adds a fixed-size striped host gate so concurrent image requests are
serialized without retaining an unbounded attacker-controlled hostname
map.

**desktop/src-tauri/src/commands/link_preview_tests.rs**
Moves native link-preview tests into a dedicated module and covers the
user-paced metadata contract, bounded one-shot cooldown behavior, and
gate release while a rate-limited request sleeps—including a different
host sharing the same bounded gate stripe.

**desktop/src-tauri/src/commands/link_preview_youtube.rs**
Removes the thumbnail fetch deadline so YouTube previews follow the same
composer lifecycle contract while using the shared bounded transport.

**desktop/src/shared/lib/useResolvedLinkPreviews.ts**
Adds development-only metadata outcome diagnostics with elapsed time and
image/fallback state, without logging encoded image payloads.

</details>

### Reproduction steps

1. Open the desktop composer and paste several GitHub pull request links
whose OpenGraph images share a host.
2. Observe that image requests are paced by host instead of racing, and
slow-but-progressing preview work remains pending rather than
immediately becoming a completed favicon fallback.
3. Send while preview work is still pending and confirm **Preparing link
preview** remains bounded by the existing post-Send budget.
4. Use **Skip** during preparation and confirm the message proceeds
immediately.
5. In a development build, inspect the console for `[link-preview]
metadata fetch completed` diagnostics containing elapsed time and image
state without base64 payloads.

### Related issue

N/A — scoped from the linked Buzz implementation room.

### Testing

At current head `dfb394aafbee537e9ffb04ad3732d08f65f30b8e`:

- Production-bound paused-time metadata regression passed through
`fetch_link_preview_metadata`; restoring the former 10-second aggregate
wrapper makes it fail at the pending assertion.
- Native link-preview module: 19/19 passed.
- `cargo check --manifest-path desktop/src-tauri/Cargo.toml` passed.
- Rust formatting and `git diff --check` passed.
- Pre-push `push-head-scope`, org safety, differential file-size,
branch-skew, and `desktop-tauri-checks` hooks passed.

At prior head `59e2dcf167b15c7a3e637ad2608008b7f9cef5f3`:

- Full Tauri Rust suite: 3,056 passed, 19 ignored; integration crates 7
+ 3 passed.
- Focused native link-preview suite: 26/26 passed.
- The pasted multi-preview workflow was exercised in the desktop app and
confirmed improved before draft publication.

---------

Signed-off-by: Taylor Ho <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@users.noreply.github.com>

* fix(acp): pace targeted overflow recovery on consumer capacity (#7325)

🤖
## Summary

When a Buzz agent falls behind on incoming messages, its connection can
make the backlog worse while trying to recover. The connection buffers
messages from the relay server until the agent is ready to process them;
if that buffer overflows, recovery previously requested history for
**every subscribed channel** and paused socket reads while sending those
requests. That adds traffic to an already overloaded connection. This
change requests history only for affected subscriptions, once the code
consuming those messages has room, with at least five seconds between
attempts.

The recovery path now:

- Combines repeated losses into one pending recovery per affected
subscription, keeping the oldest dropped timestamp so replay starts
early enough.
- Waits until at least half the consumer queue is free and the relay's
existing rate-limit delay has expired. The queue wakes recovery when
space becomes available; recovery does not periodically sample capacity
or hold queue space away from live messages.
- Attempts one subscription at a time, choosing the least recently
attempted so a busy channel cannot crowd out other channels or
membership notifications. The five-second delay starts when an attempt
finishes, including a failed write; failed writes leave recovery
pending.

Recovery is paced by available capacity, not by how often messages are
lost. This is not a larger buffer or a cutoff that abandons recovery.
Subscription identifiers, message filters, replay timestamp overlap and
duplicate filtering are unchanged; no downstream agent changes are
required.

This targets a reproducible overload **amplifier**, not every cause of
overload or every catch-up limitation. The initial live overload's cause
has not been established. Recovery remains best effort: a successful
request write is not proof of delivery, and existing history/retention
limits, bounded duplicate tracking and replay limitations still apply.
There is no exactly-once or complete catch-up guarantee. A stalled write
can still pause socket reads for the existing ten-second timeout; the
pacing bound does not cover initial subscriptions, reconnects or other
retry paths.

### Related issue

Closest related: #5014 (channel re-subscription); also #6661 (membership
reconciliation) and #6090 (relay backpressure gap signaling). This
addresses local overflow recovery scheduling, not those separate
mechanisms.

### Testing

Recorded offline comparisons against the previous behavior, with the
final implementation at `8000636f3073167c5a5107bb179c7d91160f1729`:

| Same fixture: 18 subscriptions, three overload rounds | Before | After
|
| --- | --- | --- |
| Recovery history requests | 108 | 3 |
| Ping-response delay | About 4.6 seconds | Below the measurement's 1 ms
resolution |

A separate bounded-history fixture delivered all 320 events plus
subsequent live traffic in **both** versions. Regression coverage
exercises the real socket-handling task, including intermittent consumer
capacity, fairness, failed writes and cancellation of capacity waits
before live delivery. These are synthetic results, not production
throughput measurements or evidence of a deployed cure.

The full local `RUST_TEST_THREADS=4 just ci` run passed on September 4,
2026. Earlier unsuccessful local runs remain part of the validation
history. The [recorded validation evidence and separate desktop
follow-up](https://github.com/block/buzz/pull/7325#issuecomment-5540592398)
preserve the original desktop mock-history scroll failure, its passing
rerun and the remaining investigation. That desktop path does not run
the agent connection code; neither this repair nor the passing rerun
fixes the observed scroll problem.

---------

Signed-off-by: Logan Johnson <[email protected]>

* fix(mobile): render push notification sender identity as npub (#7494)

🤖

## Summary

When an iOS push notification comes from someone the app has no cached
name for, the notification title showed the first characters of the
sender's raw public key — for example `aa4fc866…`. That fragment is
unreadable and doesn't match how the same person appears anywhere else
in Buzz. This PR changes that title to the compact form of the sender's
npub (npub is the human-readable encoding of a Nostr public key): first
8 and last 4 characters — for example `npub14f8…9nsy`, the same identity
shape used across the desktop and mobile apps.

- Unnamed senders: raw hex fragment → compact npub.
- Named senders: unchanged — a sender the app has a display name for
still titles the notification with that name.
- Unverifiable sender identities (malformed keys, or lookalike strings
that are not literal 64-hex-digit keys) now render a neutral "Someone"
instead of partial raw key material.
- Everything else about the notification is unchanged: body text,
subtitle, thread matching and grouping, deep-link navigation, thread
identifiers, and the internal hex public key the resolver matches on.

The native iOS notification-service package (`BuzzPushKit`) gains a
minimal in-house bech32 codec (bech32 is the checksummed string encoding
npubs use) — checksum-validated, 32-byte keys only, and no new external
dependency. The hex input branch accepts exactly a 64 ASCII hex digit
key before any parsing, so strings that merely parse like hex (for
example a run of `+a` pairs) cannot become a displayed identity; this is
input validation for presentation. Event signature verification is
untouched.

### Related issue

Fixes: N/A. Searched existing issues/PRs for push-notification npub
identity — closest related: none found.

### Testing

At head `3e3f2813b8864b76257ccb50dea3a4b31fa4de0d` (base
`44316ff72f5f7de014c66b01cbf534298a70c249`; 4 files, +321/−4):

- CI `Mobile Swift` lane, at this exact head — all passed: `swift test`
(73 tests, 0 failures), the SwiftPM debug and release builds of
`mobile/ios/BuzzPushKit`, and the unsigned iOS release build.
- Test coverage: npub encoding cross-checked against independent
nostr-rs/NIP-19 vectors; rejection of bad checksums, mixed case, wrong
lengths, invalid alphabet, padding, and non-32-byte payloads; resolver
boundary matrix — hex/npub/invalid sender keys render compact npub or
"Someone" while body, subtitle, sender key, and thread identifier pass
through; named senders keep cached display names.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): shared npub identity foundation (canonicalNpub, PubKey gate, strict parser) (#7488)

🤖
## Summary

Identity keys in the desktop app are displayed as raw 64-character hex.
A person's key shows up as something like `953d3363…` — unreadable,
impossible to recognize as the same identity on another screen, and a
hazard when copied by hand. Nostr (the protocol Buzz runs on) has a
human-readable spelling for identity keys — the `npub1…` form — but the
desktop app did not use it consistently.

This is the foundation of the desktop npub changes: it adds the shared
pieces every identity surface builds on, and two follow-up slices stack
directly on this branch — #7489 converts the identity controls (profile,
settings, allowlist, workflow key fields) and #7495 converts the
everyday display surfaces (mentions, member lists, sidebar, and other
name fallbacks).

After this change:

- The shared identity widget shows the compact npub form —
`npub1j57...fjmv` — instead of a hex prefix, everywhere it renders (for
example the owned-agent public-key row on a profile). Copying it puts
the full npub on the clipboard.
- Copy is a real interaction, verified end-to-end: both popover variants
put the exact canonical npub on the actual clipboard — never the raw hex
the popover also lists, never a truncation — and a portaled popover's
clicks no longer steal focus from the new-DM To-field mid-copy. Pointer
copy, a natural Space-then-Enter path, and inner/outer Escape are
covered.
- Anything that isn't a valid identity key fails neutrally: short or
corrupt values — including degenerate values that technically encode to
a checksum-valid npub but aren't real identity keys — show "Unavailable"
with no copy button, instead of a misleading value.
- Both valid npub spellings display: all-lowercase `npub1…` and
all-uppercase `NPUB1…` (Bech32, npub's encoding, permits either casing)
both render the same canonical lowercase npub. Mixed case is rejected by
the display path as written — `canonicalNpub` and the widget don't
case-normalize input — while input parsing (`parsePubkeyInput`) keeps
its trim-and-lowercase normalization and accepts mixed-case npubs; both
paths require the decoded payload to be exactly a 64-character identity
key.
- Identity-key input is strict on payload: an npub whose decoded payload
isn't exactly a 64-character identity key is rejected, matching the
validation the app's Rust side already applies to agent allowlists.

Intentional scope boundary: only surfaces that render through the shared
widget change here. Outer profile copy, settings identity cards, the
respond-to allowlist, and workflow key fields still show hex — they move
to npub in the controls follow-up (#7489). Nothing else changes identity
representation: display names, private keys, event IDs, and the hex the
app stores, sends, and matches internally are untouched; only the
user-facing spelling of an identity key changes.

## Details

- `desktop/src/shared/lib/pubkey.ts` — `canonicalNpub()`: strict
canonical full-npub helper (64-char hex in any case, or a
checksum-validated npub, returns the canonical npub; anything else
returns `null`); `truncateNpub()`: the compact display form; existing
exports unchanged.
- `desktop/src/shared/ui/PubKey.tsx` — the shared widget's identity gate
validates through `canonicalNpub`; the popover copies the npub only.
- `desktop/src/shared/lib/nostrUtils.ts` — `parsePubkeyInput` rejects
npubs whose payload is not exactly a 64-character identity key.
- `desktop/src/features/messages/ui/NewMessageScreen.tsx` — the To-field
focuses its search input only for clicks that land inside the field
itself, so portaled recipient popovers keep their focus while open (a
popover click previously dismissed it mid-copy).
- Unit suites cover the helper, widget, and parser (including the
degenerate-encode and uppercase regressions); the e2e specs that render
these rows assert the npub display.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for npub identity display —
no existing match.
- Stack: #7489 is based on this branch and builds on these primitives;
it does not stand alone on main.

### Testing

At head `b3310c248` (base: main `44316ff72`; 12 files, +440/−39):

- Focused unit suites (pubkey, PubKey, parsePubkeyInput): 20/20 green;
mutation-checked — removing the decoded-length predicate fails the
short/empty checksum-valid-npub assertions in `canonicalNpub` and the
widget, and a wrong-identity clipboard value fails the new copy
assertions.
- `pnpm typecheck` and `pnpm check`: pass; full desktop unit suite
6459/6459 at this exact head.
- Targeted e2e at this exact head: 8/8 across the two specs that own the
clipboard flows — `agent-access-warning.spec.ts` (compact variant,
agent-access owner hint) and `pubkey-display-screenshots.spec.ts` (full
variant, new-DM recipient verification: pointer copy, popover surviving
the copy, inner/outer Escape, Space-then-Enter).
- No Rust-side or build files change in this PR, so those results are
unaffected.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* test(mobile): match SDK font filename case

Signed-off-by: Michael Feth <[email protected]>

* fix(desktop): npub identity displays for mention, member, and workflow surfaces (#7495)

🤖
## Summary

Every Buzz account is identified by a long public key. Before this
change, when someone had no display name, surfaces fell back to
inconsistent labels — mostly raw hex fragments like `abcd1234…wxyz`,
sometimes a generic role label with no key — so the same person looked
different from surface to surface, and nothing looked like an npub
address. This PR applies the npub identity foundation from #7488 to the
everyday surfaces: a person without a display name now falls back to the
same compact npub everywhere — `npub1xxxx…yyyy`, the human-readable
spelling of their public key (first 8 + last 4 characters of the full
npub) — across messages and mentions, reactions, huddles, member and
participant lists, the sidebar and channel activity, search, projects,
tray, notifications, and workflow surfaces.

- **Mentions and messages**: key-only mention chips render the compact
npub. Pasting a copied mention back still re-binds it byte-exactly to
the identity it declares, for both the new npub chips and legacy
hex-truncated chips copied by older clients — wrong, missing, or
tampered key qualification is rejected instead of silently degrading to
plain text.
- **Reactions and huddles**: huddle reaction events and the huddle
roster/participants render the compact npub for unnamed participants;
workflow reaction triggers describe authors with the same form.
- **Members and sidebar**: channel and community member lists,
add-member results and invites, the members sidebar, the
channel-activity popover, search, projects (assignees/reviewers/PR
panels), the tray menu, and desktop notifications all fall back to the
compact npub; titles and aria labels keep the machine-readable full
labels.
- **Profile labels**: panel/popover display names and owner handles fall
back to the compact npub (never raw hex) when there is no name;
linked-event (nevent) message metadata shows the npub-shaped author
fallback while the event lookup and event IDs are unchanged.
- **Workflows**: author-picker secondary labels, step destination keys,
and trigger-author references render compact npubs; event and blob IDs
keep their existing hex compacts (they are not identities).
- **Avatars stay distinct**: fallback avatars for key-only identities
derive initials from the key's tail, so prefixed role labels like
"Participant npub1…" no longer collapse every unnamed participant onto
the same initials; people with names keep their name initials.

Preserved exactly: display names and distinct avatars, internal hex keys
(storage/API forms unchanged), clipboard identity roundtrips, event/blob
ID compaction, private keys (no nsec path is touched), and nevent link
handling.

Scope: this PR changes what identity labels **display**, not identity
controls — profile/settings copy controls, the respond-to allowlist,
workflow key fields, and agent dialogs are the sibling slice #7489, and
the shared primitives (`canonicalNpub`, `truncateNpub`, the `<PubKey>`
gate, strict input parsing) come from the foundation #7488.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for duplicates — none found;
the related work is the npub identity stack this slice belongs to.
- Base/dependency: stacks on #7488 (foundation) — this PR does not stand
alone on main.
- #7489 is a sibling slice on the same #7488 base
(profile/agent/workflow controls), not a dependency: this PR does not
require #7489, and #7489 does not require this PR — both only require
#7488.

### Testing

At exact head `4763cbeae1dd521309755e6d61f657324cb98667` (base:
`fix/desktop-npub-identity-d1a` @
`5f3a4a8111998c8aa41ad77cf66992bd1c85343c`; 71 files, +656/−189 —
production +277/−136, test support +379/−53):

- At this head: targeted `mentions.spec.ts` (1/1), the e2e build,
typecheck, and biome — green.
- 9 changed/related unit files: 100/100 green; typecheck, e2e build,
biome, and px text/truncation checks clean; huddle-roster focused run
green; channel-activity e2e 11/11; mutation checks confirm the fallback
wiring (removing it collapses shared initials and drops fallback rows).
- Known pre-existing local e2e failures, unchanged by this PR and
reproduced identically at the upstream merge-base: huddle-transcription
voice-menu attribution (25 pass / 1 fail) and the
`workflow-local-controls` 438px caret drift. Not claimed green locally.
- Update at head `236af9e6137386737e84d3a474d6bc808a704c50` (test-only
follow-ups `1143af345` + `236af9e6`): the `workflow-local-controls`
races were fixed in the test drivers, and the 438px diff was shown to be
a stale Darwin snapshot baseline (name-row enable switch already absent
and `message_posted` already MessageSquare at recording commit
`9390e11c9`) and refreshed — the focused screenshot test, including
keyboard/caret assertions, now passes locally (twice). The full spec was
not rerun after the snapshot refresh; the huddle-transcription item
above is unchanged.

Label/copy text changes are asserted by the e2e specs (`mentions`,
`mention-recipients`, `pubkey-display-screenshots`,
`huddle-transcription`, `channel-activity-popover`,
`workflow-local-controls`) rather than new screenshots; the screenshot
spec pins the compact npub text forms.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): npub identity controls across profile, agents, and workflows (#7489)

🤖
## Summary

Building on #7488's npub foundation, this PR finishes the identity
display change for the controls where you actually manage people and
keys: profile, settings, agent access, and workflows. Everywhere in
these surfaces, an identity key shows — and copies — as its canonical
npub (npub is the human-readable encoding of a Nostr public key: the
compact `npub1j57...fjmv` form where space is tight, the full npub where
the whole key matters), and accepts npub as input.

After this change:

- Profile panel: the public-key row and the managed-by / declared-owner
copies show the full npub. If a key can't be encoded, you see
"Unavailable" with no copy button — never a raw or partial key.
- Settings: the identity card shows and copies the npub. The
hosted-communities account identity derives from the bound key
(`pubkey_hex`) — the same authority as the mismatch gate and hosted
operations — so the display can never disagree with what the app acts
on; an unusable hex falls back to a neutral label instead of rendering
the unverified server npub. The connected claim and a community's
Connect action require that same usable bound key to match the local one
— with no usable binding the card cannot claim connected or start
Connect, while the community list, linking, and delete/rebind recovery
stay available.
- Hosted create/onboarding: the account and device identity rows in the
create flow and owner onboarding derive from the same authoritative
fields (bound key / local key), with the same neutral fallback;
readiness requires a usable bound key that matches the local one.
- Respond-to allowlist (controls who may respond to an agent): entries
can be typed or pasted as hex or npub; both spellings of the same key
are recognized as one entry and dedupe. Search results, chips, and
remove buttons use the compact npub.
- Workflow key fields: to/from keys display as npubs in the form and
save back as canonical hex. Templates like `{{trigger.author}}`, roles,
and free text pass through untouched; placeholders accept both
spellings.
- Recipient and agent dialogs: the verify popover is npub-only (the
raw-hex line is gone); denied-membership screens never show a raw key.
- The Rust-side truncated display name (used for native surfaces) shows
the same compact npub, so those surfaces match the web UI.

Internal representation is unchanged: keys are still stored, sent, and
matched as canonical 64-character hex — npub is a display and input
spelling, normalized to hex at the boundary, so existing data and
integrations keep working. Bound-key usability and comparison use one
normalized form (trimmed, lowercased, 64 hex characters; npub rejected),
so padded or mixed-case spellings of the same key match. Display names,
private keys, and event IDs are untouched.

## Details

- `respondToAllowlist` / `RespondToField`: npub entries normalize to
canonical hex; cross-form dedupe; compact npub in rows and chips;
direct-add accepts npub and stores canonical hex.
- `workflowFormTypes` / `WorkflowStepCard`: hex → npub for display, npub
→ canonical hex on save; templates, roles, and free text pass through in
both directions (roundtrip-tested).
- `UserProfilePanelFields`, `ProfileSettingsCard`,
`HostedCommunitiesSettingsCard`, `MembershipDenied`,
`SelectedRecipientChip`, `AddAgentToChannelDialog`: npub display and
copy; invalid keys → "Unavailable" with no copy; hosted identity rows
derive from the bound `pubkey_hex` (create/onboarding rows from the
bound and local keys), never the unverified server npub;
connected/readiness/Connect gates use the same usable-bound-key
predicate, and the settings Connect invocation callback re-checks it
before starting.
- `src-tauri/src/commands/identity.rs`: `truncated_display_name`
compacts to the first 8 + last 4 characters of the npub (above a 12-char
threshold), mirroring `truncateNpub`.
- e2e: profile key rows and clipboard polls assert npub forms and
raw-hex suppression; the display-screenshots spec pins the npub-only
popover; hosted specs drive the real settings card, create flow, and
onboarding rows through their real providers, and the unlinked/npub-only
identity cases assert no connected claim and no Connect action.

### Related issue

- Fixes: N/A. No separate issue; the related work is the stack below.
- Stack: builds on #7488 (shared npub foundation), now merged; this PR
is rebased onto main and stands on its own.

### Testing

At head `303c90ffa` (base: main `bfc38485`; 24 files, +1125/−146):

- Focused unit suites (respondToAllowlist, workflowFormTypes,
hostedCommunityApi bound-key helpers) green; mutation-checked — dropping
allowlist canonicalization fails the dedupe case, and dropping bound-key
normalization fails the npub-in-hex and padded same-key cases.
- Full desktop unit suite 6,477/6,477, `desktop-typecheck`,
`desktop-check` (formatting fixed narrowly with `biome check --write` on
the touched files only), and a fresh E2E build at the current head; the
add-community + hosted-communities-settings specs 18/18 and onboarding
integration 69/69 on a fresh dedicated port, with focused new-case runs
4+4 covering padded same-key (ready, Connect kept — no false rebind) and
npub-in-hex (neutral label, recovery, no Connect) across the settings
card, create flow, and first-community onboarding, plus the
unlinked-account settings regression asserting Connect cannot occur.
- `cargo fmt`/clippy (both feature sets) and `cargo test identity` (71
pass) passed at the earlier full-change head; since then, the only
production changes in this PR's delta are the hosted identity display
authority and its fail-closed bound-key gating/normalization above
(base-side fixes carry #7488's receipts) — every other change is
test-only.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(mobile): standardize public-key identity display on npub (#7493)

🤖

## Summary

In the mobile app, anyone who hasn't set a display name shows up as a
raw 64-character hex key (e.g. `3a5d4f9c…`) — unreadable, and
unrecognizable as the same identity across screens. Profile and Settings
also let you copy that raw hex. Nostr public keys have a standard
readable form — `npub1…`, the same encoding other Nostr apps and our
desktop app already display. This PR makes every mobile identity surface
render npub instead:

- **Unnamed people everywhere** — message and thread authors, reactions,
typing indicators, member lists, channel details, DM headers and tiles,
inbox, search, forum cards, Pulse notes and reply context, mention
suggestions, and invite rows — now show a compact npub label: first 8 +
last 4 characters of the full npub joined by an ellipsis
(`npub1abcd…wxyz`), the same truncation desktop uses. Previously these
showed truncated raw hex.
- **DM fallback avatars and blank names** — 1:1 DM tiles and headers key
their fallback avatar to the same non-self counterpart the label names,
including self-first participant order; a self-DM keeps its
hex-key-derived initial. Blank or whitespace-only display names fall
back to the compact npub instead of rendering empty, while nonblank
authored names render verbatim (padding included).
- **Profile sheet → "Copy public key"** now copies the full canonical
npub — never raw hex. When the identity string isn't a valid public key,
the copy tile is disabled, so a malformed key never reaches the
clipboard.
- **Settings → Identity (pubkey)** displays and copies the full npub; an
invalid identity reads "Identity unavailable" with copy disabled.
- **Invalid identities never leak truncated raw hex** into the UI
anywhere — they render a neutral "Unknown identity" label.
- **Unchanged on purpose:** display names and verified handles (NIP-05 —
the `name@domain` badge) still render as before. Unnamed avatars keep
distinct per-key initials, derived from the underlying hex key rather
than the npub — otherwise every unnamed key would render the same "N"
initial. Event IDs are not public keys, so they keep their hex
truncation (in Pulse's "Replying to", the parent author shows npub while
an event-id fallback still shows hex). The nevent share link, private
keys, and internal hex storage are untouched. Inputs that accept a key
(invite/member entry) accept both hex and npub and keep working in hex
internally.

### Related issue

N/A. Searched open issues/PRs for npub identity display on mobile —
closest related: none found. Desktop's parallel npub standardization
lives in the stacked desktop PRs (#7488 foundation, #7489 controls,
#7495 display surfaces); this is the independent mobile slice (based
directly on `main`, not on those branches).

### Testing

At exact head `5a620e420a1fd57d9d8011ac26434eed32fcf765` (base: `main`
`44316ff72`; 40 files, +1,345/−154):

- Full mobile suite: 2,098 tests passing (`cd mobile && flutter test`);
`flutter analyze` clean; `dart format --set-exit-if-changed .` clean —
the same checks CI runs.
- Widget/unit coverage at production seams: compact labels and hex-keyed
avatar initials for DM headers/tiles, member rows, mention suggestions,
and Pulse reply context; DM fallback avatars keyed to the labeled
counterpart (self-first order and self-DMs); blank/whitespace
display-name npub fallback with nonblank authored labels verbatim,
including the Activity inbox sender and profile-sheet heading (each with
its own empty/whitespace production-seam regression); full-npub copy and
disabled-copy semantics in profile and settings; invalid-key
suppression; and hex↔npub input round-trips.

Verified via unit and widget tests — no device/simulator validation is
claimed.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* test(mobile): align fork identity fixtures with upstream npub labels

Upstream #7488/#7493 made the compact npub the canonical identity label
and made shortPubkey return 'Unknown identity' for non-key strings. Three
fork tests still expected raw-hex truncation or accepted invalid keys:
update fixtures/expectations to the npub contract. Behavior under test
(plain-text actor labels, digest author naming) is unchanged.

Signed-off-by: Michael Feth <[email protected]>

* fix(desktop): order unnamed roster members by full canonical npub (#7503)

🤖

## Summary

- Channel members appear in the Members sidebar. A member who has never
set a display name is listed under an abbreviated form of their public
key (npub), and the sidebar previously sorted those unnamed members by
that short label. Short labels are not unique — different keys can share
one — so the order of unnamed members could look arbitrary or unstable.
Unnamed members now sort by their full public key, so the order is
deterministic.
- When two members display the same name, the previous tiebreak was
membership order (who joined first), which is not visible to a reader
and can shift as roster data loads in. The tiebreak is now the full
public key, so identical display names always land in the same order.
- Nothing gets noisier on screen: the full key is used only for sorting,
and the sidebar still shows the compact abbreviated form. Priorities are
unchanged — authored (custom) names still outrank fallback labels, and
role/current-user grouping still applies.
- Scope is the desktop app's Members sidebar and member management: the
two existing sort comparators. Mobile and other lists in the app are
untouched.

### Related issue

Based on #7495 (introduced the abbreviated npub labels this follows up
on). The original five presentation PRs remain independently reviewable.
No closer duplicate found.

### Testing

- 6469 desktop unit tests, typecheck, and check pass.
- The 3 existing consumer-seam E2E tests still pass; a new E2E test
asserts the sidebar lists unnamed members in full-key order, with
fixture members deliberately inserted in the opposite order so incoming
membership order cannot mask the sort.
- Negative check: reverting only this change makes the new ordering
assertion fail, so it genuinely binds the new sort.
- CI has not run on this PR yet.

Buzz provenance: channel 1f0e4a3d-7e01-4efe-bb16-843b357f85c9 / task
340c3de9b27dbedb8453c0c7652220f9080d30fcc70a7c4f6e27fdd4fa378056

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): require a Codex adapter with Astra support (#7427)

## Summary

Buzz considers codex-acp 1.6.2 current because the supported adapter
floor is still 1.1.7. That adapter bundles Codex 0.148.0, so updating a
separate Codex CLI to 0.153.4 leaves managed agents on the older runtime
and unable to use GPT-6 Astra.

Raise the supported adapter floor to the published 1.10.0 release, which
depends on `@openai/codex ^0.153.3`. Existing discovery and installation
code then classifies older adapters as outdated and offers the managed
reinstall path. Update the availability and install-plan regressions to
cover the observed 1.6.2 installation and the new minimum.

This follows the existing version-floor policy. It does not
automatically update a running installation: the user must complete
Buzz’s offered adapter upgrade. Future upstream compatibility changes
may require another floor update.

### Related issue

No exact duplicate found in searches for Astra, CODEX_PATH, bundled
Codex, outdated runtime, and codex-acp 1.10. Related: #3097 raised the
older floor to 1.1.7 (already present on main); #2422 covers lost error
details for runtime mismatches. Neither resolves this version gap.

Originating conversation:
buzz://message?channel=3286cd76-f83e-4c7d-8317-10a16580744d&id=8b79a73078217222b870fff144c27e7d27bcd5a67c966869c18fe726db716898

### Testing

- Isolated npm install of codex-acp 1.10.0 resolved bundled Codex
0.153.4, with no CODEX_PATH override.
- Live macOS ACP probe: initialize protocol v1 → session/new → select
gpt-6-astra[medium] → prompt. Received `OK` and `stopReason: end_turn`;
usage metadata confirms gpt-6-astra.
- Existing adapter 1.6.2 initialized but advertised no Astra model in
the same probe.
- Desktop Rust formatting and `git diff --check` pass.
- `just desktop-tauri-test`: 3,266 passed, 20 ignored, zero failures
across the Desktop workspace and integration tests.
- Workspace and Desktop Clippy, frontend static checks, and `just
file-size-check` pass.
- Repository `just ci`: still running the remaining
mobile/build/workspace-test stages.

The installed Buzz app and managed adapter were not replaced or
restarted. The live check validates the new adapter/runtime path; a
complete packaged Desktop upgrade workflow remains untested.

Signed-off-by: Stephen DeLorme <[email protected]>

* fix(buzz-acp): report missing models without retrying (#7538)

## Summary

When an agent reports model-not-found, Buzz retries the unavailable
model and delays the failure reply until retries are exhausted. Stop
retrying this error and immediately post a threaded recovery notice. The
notice tells users to select a different model in agent settings, save,
restart the agent to apply the configuration, and re-send their request.

This adds one error-handling branch and regression coverage in
`buzz-acp`. It matches `-32002` errors containing `model not found`.
Other resource-not-found errors, such as stale sessions, retain the
existing retry behavior. Detailed error events remain available for
diagnosis. The existing restart policy is unchanged.

### Related issue

None found in existing issue/PR searches for model-not-found recovery.

### Testing

Playwright captured and visually checked the thread UI with seeded
conversation data and the exact recovery text. The check opens the
request's thread, confirms no reply before the failure, injects the
notice, and verifies the full text is visible. [Before/after
screenshots](https://github.com/block/buzz/pull/7538#issuecomment-5608196506)
show the corrected save-and-restart instructions. These are local test
captures, not a deployed provider recovery flow.

Generated with Codex

---------

Signed-off-by: Diem Nguyen <[email protected]>

* [lenny] fix(deps): sync Cargo.lock with merged Cargo.toml manifests

The upstream merge combined new dependencies from both sides but left
Cargo.lock at its pre-merge state. Docker builds run with --locked and
reject a lockfile that does not match the manifests:
  'the lock file needs to be updated but --locked was passed'

Adds only the union of both sides' new deps (buzz-acp: async-trait,
rusqlite, tempfile; buzz-persona: url). No version changes.

* [lenny] fix(mobile): dart format thread_summarize_action_test

'dart format' collapses the _message(...) call onto one line (fits in
80 cols). Fixes the mobile format CI failure.

Signed-off-by: Michael Feth <[email protected]>

* [lenny] fix(deps): bump rustls 0.23.42 -> 0.23.45 for RUSTSEC-2026-0285

TLS 1.3 handshake messages were accepted across encryption level
boundaries; patched in >= 0.23.45. Lockfile-only bump (rustls,
rustls-webpki 0.103.15, aws-lc-rs 1.18.1, aws-lc-sys 0.45.0).

Signed-off-by: Michael Feth <[email protected]>

* test(desktop): scope e2e text locators past the timeline aria-live region

The sr-only TimelineAnnouncementRegion echoes the latest message text,
so bare getByText(...) locators resolve to 2 elements (strict mode
violation) whenever the announcement matches the asserted text.

thread-unread 15-mark-read-unread-menu-single-toggle failed 3/3 on
runs 34877681776, 34877661550, 34666787614 and product/main with
'getByText("Toggle me") resolved to 2 elements'; same pattern in
thread-reply-anchor-roleplay 01/03/04 ('Nora: adding context',
'Brain: Check the anchor' ...). Scope the locators to the timeline /
thread panel so they resolve uniquely.

Local: pnpm build:e2e + playwright smoke on both specs = 17 passed.
Signed-off-by: Lenny <[email protected]>

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Signed-off-by: Salman Mohammed <[email protected]>
Signed-off-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>
Signed-off-by: Michael Neale <[email protected]>
Signed-off-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Signed-off-by: Alessandro Joabar <[email protected]>
Signed-off-by: Taylor Ho <[email protected]>
Signed-off-by: Logan Johnson <[email protected]>
Signed-off-by: Michael Feth <[email protected]>
Signed-off-by: Stephen DeLorme <[email protected]>
Signed-off-by: Diem Nguyen <[email protected]>
Signed-off-by: Lenny <[email protected]>
Co-authored-by: TolgaCinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>
Co-authored-by: Salman Mohammed <[email protected]>
Co-authored-by: Bradley Axen <[email protected]>
Co-authored-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>
Co-authored-by: Mic Neale <[email protected]>
Co-authored-by: Michael Neale <[email protected]>
Co-authored-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Co-authored-by: Mic Neale <[email protected]>
Co-authored-by: Alessandro Joabar <[email protected]>
Co-authored-by: Taylor Ho <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@users.noreply.github.com>
Co-authored-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>
Co-authored-by: Stephen DeLorme <[email protected]>
Co-authored-by: Diem Nguyen <[email protected]>
Co-authored-by: Lenny <[email protected]>

…CP session scope, thread-context dedup) (#64)

* fix(mobile): style inline code with the app mono face (#6631)

## Summary

Inline code on mobile renders as **bold body text on a faint background
wash** — no monospace face, no chip, and it cannot wrap. #5257 diagnosed
this as a missing `highlightBuilder`.

That is no longer the right fix. `gpt_markdown` 1.2.0 deprecates
`highlightBuilder` (removal in 2.0.0), renders inline code as a real
chip, and adds `InlineCodeStyle` for restyling it. The package author
confirmed this on the issue. So this PR is an upgrade — 1.1.6 → 1.2.1 —
plus one theme declaration, rather than the builder the issue originally
asked for.

**Where the style is declared.** `GptMarkdownThemeData` goes in
`AppTheme._buildTheme`, which both `light()` and `dark()` call. That
reaches all four `GptMarkdown` call sites — `message_content`,
`transcript_item_widget`, `token_pill`, `custom_emoji_render` — so the
style is stated once instead of per widget. A widget-level
`inlineCodeStyle` would have covered channel messages only, leaving the
other three on the package's defaults.

**What is declared.** Face, size, ink, chip fill and outline — not the
face alone. A face name on its own leaves the rest on the package's
defaults, which put inline code at 14.1sp beside a fenced block's 13, on
a neutral `onSurface` tint rather than the app's code surface. In dark
that tint is *lighter* than the surface, while every other code surface
in the app is recessed, so the chip read as a different kind of object.
All of it now comes from one `CodeStyle` declaration that the fenced
block reads from too, so the two cannot be edited apart.

**Three adaptations the upgrade requires.** Each was found by running
the gate, not by reading the changelog:

1. **`imageBuilder` widened** to `(context, url, width, height)`. This
is a hard compile error, and it is **not listed in the package's
migration guide**, which states "nothing here stops code compiling".
Worth reporting upstream.
2. **`autolink` now defaults to `true`.** `normalizeBareLinks()` already
rewrites bare URLs into Markdown links before rendering, so both would
run. `message_content` opts out with `autolink: false` to keep current
behaviour exactly. The migration guide argues for dropping the
pre-processor instead — a better fix, but a behavioural change that
belongs in its own PR.
3. **`gpt_markdown.dart` now re-exports `markdown_config.dart`**, making
two direct imports redundant. `flutter analyze` reports `No issues
found!` on 1.1.6 and flags both on 1.2.1, so these warnings are new, not
pre-existing.

**Deliberately out of scope.** The three non-message call sites now
autolink bare URLs, since only `message_content` has a pre-processor to
collide with. Custom inline components (`_MentionMd`, `CustomEmojiMd`,
`_ChannelLinkMd`) could additionally declare `allScopesExceptLinkLabel`
— 1.2.0 offers it as the fix for a `WidgetSpan` chip going blank inside
a link label on iOS — but current behaviour is unchanged without it, so
that stays a separate change.

### Related issue

Fixes #5257

Duplicate scan: searched `gpt_markdown`, `inline code mobile`,
`highlightBuilder` and `InlineCodeStyle` across both PRs and issues. No
open PR touches inline code styling. #6135 (link labels) and #6166 (text
selection) also touch mobile Markdown but address different defects.

### Testing

Full gate, `just ci` — exit 0:

| Stage | Result |
|---|---|
| Rust (33 suites) | 4768 passed, 0 failed |
| Desktop | 5799 passed, 0 failed |
| Mobile | **2011 passed**, 0 failed |
| `flutter analyze` | `No issues found!` |
| Desktop + web build | ok |

Run on the branch with `main` merged in, so these numbers match what CI
builds.

**New regression test** — `renders inline code in the app code style`.
It resolves the `CodeTextSpan` the package tags inline code with, which
carries both the resolved `TextStyle` and the colours the chip behind it
is painted with, so face, size, ink, fill and outline are all asserted
rather than a widget's presence. It is negative-controlled: reverting
only the theme declaration fails it with

```text
Expected: a numeric value within <0.001> of <13.0>
  Actual: <14.1>
```

and dropping the declaration entirely falls back to
`packages/gpt_markdown/JetBrainsMono` — so the test measures the real
thing, and it would catch a future regression that silently drops the
theme extension.

The test passes `baseStyle: messageBodyTextStyle`, the style the message
surfaces actually use; the widget's own fallback is the smaller
`bodyMedium`, which would move the expected size.

The test finds paragraphs with `find.byWidgetPredicate((widget) =>
widget is RichText)`, not `find.byType(RichText)`: inline code renders
through `BidiRichText`, a `RichText` subclass, and `byType` matches
exact runtime types.

That is a hazard for any test that reads text back out of a paragraph,
and one landed after this branch was cut:
`message_content_custom_emoji_test.dart` arrived with #6996 and its
`code keeps literal emoji while adjacent known tokens render` case reads
a code span through `find.byType(RichText)`. It passes on `main` and
fails on the merge result, which is what CI builds, so it went red only
once CI was authorized. It now uses the same predicate. The two other
`byType(RichText)` call sites — the rest of that file and
`message_author_meta_test.dart` — were re-run and pass: their content
carries no code span, so the exact type still matches. They were left
alone.

### Screenshots

Rendered through the real `MessageContent` widget with the app's own
fonts loaded, at 390pt wide, 3x DPR. Sample text: ``Set `BUZZ_RELAY_URL`
before launch, then run `just mobile-test` to verify.``

| | Before (1.1.6) | After (1.2.1) |
|---|---|---|
| Light |
![before-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-light.png)
|
![after-inline-code-light](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-light.png)
|
| Dark |
![before-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/2d2d846291416d9b32d3fb9cfead950bcc4fe123/pr-6631--before-inline-code-dark.png)
|
![after-inline-code-dark](https://raw.githubusercontent.com/TolgaCinisli/buzz/f230b95c7260a32bd5d76b1ac42130720a168521/pr-6631--after-inline-code-dark.png)
|

Before: bold Inter on a flat wash, no chip edge, and `just mobile-test`
breaks across the line with the wash simply ending. After: Geist Mono in
a bordered, rounded chip, and the wrapped fragment gets its own chip on
each line.

---------

Signed-off-by: Tolga Cinisli <[email protected]>
Co-authored-by: Tolga Cinisli <[email protected]>

* fix(buzz-acp): wake held ACP threads and fence forked sessions (#7340)

## Summary

Adds an independent deadline wakeup so held thread work dispatches after
its 10-second bound even when the relay loop is otherwise quiet. Fences
session ownership by generation so a worker returning after a fork
cannot make an older provider session claimable again.

This follows up on the two post-merge findings from
[#7337](https://github.com/block/buzz/pull/7337#pullrequestreview-5116329341).

### Related issue

Follow-up to #7337.

### Testing

- `cargo test -p buzz-acp`
- `cargo clippy -p buzz-acp --all-targets -- -D warnings`
- Pre-push file-size, differential Rust test, and desktop Tauri gates

No UI changes.

---
**Update Sep 4, 15:35:** Addressed both Codex review findings.
- Queue-cap eviction now prunes orphaned hold deadlines.
- An expired hold stays expired until a worker is successfully claimed.
- Hold timers remain disabled while every worker is busy; worker return
wakes dispatch directly.
- Added regressions for queue eviction and pool exhaustion.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

* fix(agent): route GPT-5+ model-service FQNs to Responses (#7358)

## Summary
Route Databricks Unity Catalog model services to OpenAI Responses when
their service name matches GPT-5 or newer. These models can reject tools
plus reasoning on Chat Completions.

Match only the service component, using the existing family-token
boundaries and a numeric major version. Catalog and schema names cannot
select the protocol. Keep neutral effort capabilities and the full model
ID unchanged; other services still use MLflow Chat Completions.

Keep the Rust and desktop resolvers in sync, add shared boundary cases
and a captured-HTTP regression for completion and summarization, and
update the documented FQN rule.

### Related issue
No duplicate found in searches for “FQN responses” PRs or “astra”
issues. Related: #6918 introduced Unity Catalog discovery.

Originating conversation:
buzz://message?channel=0b881928-a3a6-4c01-b981-8e64268f01ce&id=770949343bc96a9ed88acd90a1b37d358a0efc52c79237d0fdb491ce02b8d4ed

### Testing
No live Databricks inference test. The gateway must accept the full
model-service ID on its OpenAI Responses route; this remains the
integration risk.

The local `just ci` attempt exceeded its five-minute deadline during
`mobile-check`, so the full repository gate was not completed. All
push-hook checks passed.

Generated with Codex

Signed-off-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>
Co-authored-by: Fizz <400e8babadcee6a7f420103f10a2849d84c4a9c71d5bd04f3948c814216648a3@buzz.block.builderlab.xyz>

* feat(mesh): upgrade to mesh-llm 0.76.0-rc8 and recommend Qwen3.8 27B (#6189)

Upgrades Buzz's mesh-llm dependencies through the released `v0.76.0-rc8`
tag (`2040765d`), including the Qwen3.8 curated recommendation and rc8's
scheduler/runtime improvements.

**Scope note:** the earlier open-relay/unenforced-admission mode has
been removed from this PR at Mic's direction — it is not a product mode
we want. Mesh admission remains roster/allowlist driven, exactly as on
`main`: on a relay with no NIP-43 membership snapshot the mesh runs
self-only. No NIP-11 mode probing, no mode-transition restarts. A future
perimeter/admission strategy for open relays will be designed
separately.

This PR also:
- seeds `BUZZ_AGENT_LLM_TIMEOUT_SECS=660` for mesh agents, above
MeshLLM's 600-second backend timeout;
- makes `desktop-tauri-clippy` lint both default and `mesh-llm` cfg
graphs;
- runs the feature-enabled desktop test suite in CI;
- recommends Qwen3.8 27B Q4_K_M for 64 GB-and-larger machines, then
ladders down through Gemma 4 E4B and Qwen 9B for smaller machines;
- keeps stored shared-compute `auto` translated to MeshLLM's supported
wire model `mesh`.

RC8 verification:
- `just ci` passed locally at
`92ecc7ec933bdd4df804cc9f28a2b51efa5313c5`.
- Pre-push differential gates passed, including both desktop Tauri cfg
graphs and package tests.
- A prior isolated runtime smoke used the RC8 binary's OpenAI endpoint
for a Buzz-shaped system/user/tool/tool-result/final-response loop; all
assertions passed and the isolated process was shut down.

Perf previously measured on M5 Metal, Qwen3.8-27B-Q4_K_M: TTFT 0.22–0.32
s, ~25 tok/s streaming; agent-shaped turns ~1 s to first token after the
first (prefix cache).

---------

Signed-off-by: Michael Neale <[email protected]>
Signed-off-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Signed-off-by: Alessandro Joabar <[email protected]>
Co-authored-by: Michael Neale <[email protected]>
Co-authored-by: Jimmy <1fe240cd1a8cf775f6f3060f115e5a303181f3abf28ad4cb0c2515f4a02b36a8@meshllm.communities.buzz.xyz>
Co-authored-by: Mic Neale <[email protected]>
Co-authored-by: Alessandro Joabar <[email protected]>

* fix(link-preview): keep composer fetches user-paced (#7211)

**Category:** fix
**User Impact:** Link previews can keep loading while a message is being
composed, while sending still has a finite escape hatch and stalled
network transports cannot occupy preview slots forever.

**Problem:** Native metadata and image deadlines could collapse slow
previews into fallback cards while the user was still composing, and a
shared image-host cooldown made pasted batches fail inconsistently after
one rate limit. **Solution:** Keep preview resolution user-paced with no
aggregate request deadline, bound transport inactivity (15s DNS/connect,
30s idle read), serialize image requests by host, and allow at most one
server-directed cooldown wait of up to 30s across an image fetch and its
redirects. The existing bounded post-Send preparation and immediate Skip
paths remain unchanged.

<details>
<summary>File changes</summary>

**desktop/src-tauri/src/commands/link_preview.rs**
Removes aggregate native deadlines so composer metadata work can
complete at the user's pace, while retaining DNS/connect/idle-read
liveness bounds. Adds bounded host-paced image request coordination that
releases its gate during cooldown, waits inline at most once for at most
30 seconds, and cannot renew that wait through redirects or the outer
transient retry. Same-host image and favicon requests remain
deliberately serialized to align with host rate limits.

**desktop/src-tauri/src/commands/link_preview_rate_limit.rs**
Adds a fixed-size striped host gate so concurrent image requests are
serialized without retaining an unbounded attacker-controlled hostname
map.

**desktop/src-tauri/src/commands/link_preview_tests.rs**
Moves native link-preview tests into a dedicated module and covers the
user-paced metadata contract, bounded one-shot cooldown behavior, and
gate release while a rate-limited request sleeps—including a different
host sharing the same bounded gate stripe.

**desktop/src-tauri/src/commands/link_preview_youtube.rs**
Removes the thumbnail fetch deadline so YouTube previews follow the same
composer lifecycle contract while using the shared bounded transport.

**desktop/src/shared/lib/useResolvedLinkPreviews.ts**
Adds development-only metadata outcome diagnostics with elapsed time and
image/fallback state, without logging encoded image payloads.

</details>

### Reproduction steps

1. Open the desktop composer and paste several GitHub pull request links
whose OpenGraph images share a host.
2. Observe that image requests are paced by host instead of racing, and
slow-but-progressing preview work remains pending rather than
immediately becoming a completed favicon fallback.
3. Send while preview work is still pending and confirm **Preparing link
preview** remains bounded by the existing post-Send budget.
4. Use **Skip** during preparation and confirm the message proceeds
immediately.
5. In a development build, inspect the console for `[link-preview]
metadata fetch completed` diagnostics containing elapsed time and image
state without base64 payloads.

### Related issue

N/A — scoped from the linked Buzz implementation room.

### Testing

At current head `dfb394aafbee537e9ffb04ad3732d08f65f30b8e`:

- Production-bound paused-time metadata regression passed through
`fetch_link_preview_metadata`; restoring the former 10-second aggregate
wrapper makes it fail at the pending assertion.
- Native link-preview module: 19/19 passed.
- `cargo check --manifest-path desktop/src-tauri/Cargo.toml` passed.
- Rust formatting and `git diff --check` passed.
- Pre-push `push-head-scope`, org safety, differential file-size,
branch-skew, and `desktop-tauri-checks` hooks passed.

At prior head `59e2dcf167b15c7a3e637ad2608008b7f9cef5f3`:

- Full Tauri Rust suite: 3,056 passed, 19 ignored; integration crates 7
+ 3 passed.
- Focused native link-preview suite: 26/26 passed.
- The pasted multi-preview workflow was exercised in the desktop app and
confirmed improved before draft publication.

---------

Signed-off-by: Taylor Ho <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@users.noreply.github.com>

* fix(acp): pace targeted overflow recovery on consumer capacity (#7325)

🤖
## Summary

When a Buzz agent falls behind on incoming messages, its connection can
make the backlog worse while trying to recover. The connection buffers
messages from the relay server until the agent is ready to process them;
if that buffer overflows, recovery previously requested history for
**every subscribed channel** and paused socket reads while sending those
requests. That adds traffic to an already overloaded connection. This
change requests history only for affected subscriptions, once the code
consuming those messages has room, with at least five seconds between
attempts.

The recovery path now:

- Combines repeated losses into one pending recovery per affected
subscription, keeping the oldest dropped timestamp so replay starts
early enough.
- Waits until at least half the consumer queue is free and the relay's
existing rate-limit delay has expired. The queue wakes recovery when
space becomes available; recovery does not periodically sample capacity
or hold queue space away from live messages.
- Attempts one subscription at a time, choosing the least recently
attempted so a busy channel cannot crowd out other channels or
membership notifications. The five-second delay starts when an attempt
finishes, including a failed write; failed writes leave recovery
pending.

Recovery is paced by available capacity, not by how often messages are
lost. This is not a larger buffer or a cutoff that abandons recovery.
Subscription identifiers, message filters, replay timestamp overlap and
duplicate filtering are unchanged; no downstream agent changes are
required.

This targets a reproducible overload **amplifier**, not every cause of
overload or every catch-up limitation. The initial live overload's cause
has not been established. Recovery remains best effort: a successful
request write is not proof of delivery, and existing history/retention
limits, bounded duplicate tracking and replay limitations still apply.
There is no exactly-once or complete catch-up guarantee. A stalled write
can still pause socket reads for the existing ten-second timeout; the
pacing bound does not cover initial subscriptions, reconnects or other
retry paths.

### Related issue

Closest related: #5014 (channel re-subscription); also #6661 (membership
reconciliation) and #6090 (relay backpressure gap signaling). This
addresses local overflow recovery scheduling, not those separate
mechanisms.

### Testing

Recorded offline comparisons against the previous behavior, with the
final implementation at `8000636f3073167c5a5107bb179c7d91160f1729`:

| Same fixture: 18 subscriptions, three overload rounds | Before | After
|
| --- | --- | --- |
| Recovery history requests | 108 | 3 |
| Ping-response delay | About 4.6 seconds | Below the measurement's 1 ms
resolution |

A separate bounded-history fixture delivered all 320 events plus
subsequent live traffic in **both** versions. Regression coverage
exercises the real socket-handling task, including intermittent consumer
capacity, fairness, failed writes and cancellation of capacity waits
before live delivery. These are synthetic results, not production
throughput measurements or evidence of a deployed cure.

The full local `RUST_TEST_THREADS=4 just ci` run passed on September 4,
2026. Earlier unsuccessful local runs remain part of the validation
history. The [recorded validation evidence and separate desktop
follow-up](https://github.com/block/buzz/pull/7325#issuecomment-5540592398)
preserve the original desktop mock-history scroll failure, its passing
rerun and the remaining investigation. That desktop path does not run
the agent connection code; neither this repair nor the passing rerun
fixes the observed scroll problem.

---------

Signed-off-by: Logan Johnson <[email protected]>

* fix(mobile): render push notification sender identity as npub (#7494)

🤖

## Summary

When an iOS push notification comes from someone the app has no cached
name for, the notification title showed the first characters of the
sender's raw public key — for example `aa4fc866…`. That fragment is
unreadable and doesn't match how the same person appears anywhere else
in Buzz. This PR changes that title to the compact form of the sender's
npub (npub is the human-readable encoding of a Nostr public key): first
8 and last 4 characters — for example `npub14f8…9nsy`, the same identity
shape used across the desktop and mobile apps.

- Unnamed senders: raw hex fragment → compact npub.
- Named senders: unchanged — a sender the app has a display name for
still titles the notification with that name.
- Unverifiable sender identities (malformed keys, or lookalike strings
that are not literal 64-hex-digit keys) now render a neutral "Someone"
instead of partial raw key material.
- Everything else about the notification is unchanged: body text,
subtitle, thread matching and grouping, deep-link navigation, thread
identifiers, and the internal hex public key the resolver matches on.

The native iOS notification-service package (`BuzzPushKit`) gains a
minimal in-house bech32 codec (bech32 is the checksummed string encoding
npubs use) — checksum-validated, 32-byte keys only, and no new external
dependency. The hex input branch accepts exactly a 64 ASCII hex digit
key before any parsing, so strings that merely parse like hex (for
example a run of `+a` pairs) cannot become a displayed identity; this is
input validation for presentation. Event signature verification is
untouched.

### Related issue

Fixes: N/A. Searched existing issues/PRs for push-notification npub
identity — closest related: none found.

### Testing

At head `3e3f2813b8864b76257ccb50dea3a4b31fa4de0d` (base
`44316ff72f5f7de014c66b01cbf534298a70c249`; 4 files, +321/−4):

- CI `Mobile Swift` lane, at this exact head — all passed: `swift test`
(73 tests, 0 failures), the SwiftPM debug and release builds of
`mobile/ios/BuzzPushKit`, and the unsigned iOS release build.
- Test coverage: npub encoding cross-checked against independent
nostr-rs/NIP-19 vectors; rejection of bad checksums, mixed case, wrong
lengths, invalid alphabet, padding, and non-32-byte payloads; resolver
boundary matrix — hex/npub/invalid sender keys render compact npub or
"Someone" while body, subtitle, sender key, and thread identifier pass
through; named senders keep cached display names.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): shared npub identity foundation (canonicalNpub, PubKey gate, strict parser) (#7488)

🤖
## Summary

Identity keys in the desktop app are displayed as raw 64-character hex.
A person's key shows up as something like `953d3363…` — unreadable,
impossible to recognize as the same identity on another screen, and a
hazard when copied by hand. Nostr (the protocol Buzz runs on) has a
human-readable spelling for identity keys — the `npub1…` form — but the
desktop app did not use it consistently.

This is the foundation of the desktop npub changes: it adds the shared
pieces every identity surface builds on, and two follow-up slices stack
directly on this branch — #7489 converts the identity controls (profile,
settings, allowlist, workflow key fields) and #7495 converts the
everyday display surfaces (mentions, member lists, sidebar, and other
name fallbacks).

After this change:

- The shared identity widget shows the compact npub form —
`npub1j57...fjmv` — instead of a hex prefix, everywhere it renders (for
example the owned-agent public-key row on a profile). Copying it puts
the full npub on the clipboard.
- Copy is a real interaction, verified end-to-end: both popover variants
put the exact canonical npub on the actual clipboard — never the raw hex
the popover also lists, never a truncation — and a portaled popover's
clicks no longer steal focus from the new-DM To-field mid-copy. Pointer
copy, a natural Space-then-Enter path, and inner/outer Escape are
covered.
- Anything that isn't a valid identity key fails neutrally: short or
corrupt values — including degenerate values that technically encode to
a checksum-valid npub but aren't real identity keys — show "Unavailable"
with no copy button, instead of a misleading value.
- Both valid npub spellings display: all-lowercase `npub1…` and
all-uppercase `NPUB1…` (Bech32, npub's encoding, permits either casing)
both render the same canonical lowercase npub. Mixed case is rejected by
the display path as written — `canonicalNpub` and the widget don't
case-normalize input — while input parsing (`parsePubkeyInput`) keeps
its trim-and-lowercase normalization and accepts mixed-case npubs; both
paths require the decoded payload to be exactly a 64-character identity
key.
- Identity-key input is strict on payload: an npub whose decoded payload
isn't exactly a 64-character identity key is rejected, matching the
validation the app's Rust side already applies to agent allowlists.

Intentional scope boundary: only surfaces that render through the shared
widget change here. Outer profile copy, settings identity cards, the
respond-to allowlist, and workflow key fields still show hex — they move
to npub in the controls follow-up (#7489). Nothing else changes identity
representation: display names, private keys, event IDs, and the hex the
app stores, sends, and matches internally are untouched; only the
user-facing spelling of an identity key changes.

## Details

- `desktop/src/shared/lib/pubkey.ts` — `canonicalNpub()`: strict
canonical full-npub helper (64-char hex in any case, or a
checksum-validated npub, returns the canonical npub; anything else
returns `null`); `truncateNpub()`: the compact display form; existing
exports unchanged.
- `desktop/src/shared/ui/PubKey.tsx` — the shared widget's identity gate
validates through `canonicalNpub`; the popover copies the npub only.
- `desktop/src/shared/lib/nostrUtils.ts` — `parsePubkeyInput` rejects
npubs whose payload is not exactly a 64-character identity key.
- `desktop/src/features/messages/ui/NewMessageScreen.tsx` — the To-field
focuses its search input only for clicks that land inside the field
itself, so portaled recipient popovers keep their focus while open (a
popover click previously dismissed it mid-copy).
- Unit suites cover the helper, widget, and parser (including the
degenerate-encode and uppercase regressions); the e2e specs that render
these rows assert the npub display.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for npub identity display —
no existing match.
- Stack: #7489 is based on this branch and builds on these primitives;
it does not stand alone on main.

### Testing

At head `b3310c248` (base: main `44316ff72`; 12 files, +440/−39):

- Focused unit suites (pubkey, PubKey, parsePubkeyInput): 20/20 green;
mutation-checked — removing the decoded-length predicate fails the
short/empty checksum-valid-npub assertions in `canonicalNpub` and the
widget, and a wrong-identity clipboard value fails the new copy
assertions.
- `pnpm typecheck` and `pnpm check`: pass; full desktop unit suite
6459/6459 at this exact head.
- Targeted e2e at this exact head: 8/8 across the two specs that own the
clipboard flows — `agent-access-warning.spec.ts` (compact variant,
agent-access owner hint) and `pubkey-display-screenshots.spec.ts` (full
variant, new-DM recipient verification: pointer copy, popover surviving
the copy, inner/outer Escape, Space-then-Enter).
- No Rust-side or build files change in this PR, so those results are
unaffected.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): npub identity displays for mention, member, and workflow surfaces (#7495)

🤖
## Summary

Every Buzz account is identified by a long public key. Before this
change, when someone had no display name, surfaces fell back to
inconsistent labels — mostly raw hex fragments like `abcd1234…wxyz`,
sometimes a generic role label with no key — so the same person looked
different from surface to surface, and nothing looked like an npub
address. This PR applies the npub identity foundation from #7488 to the
everyday surfaces: a person without a display name now falls back to the
same compact npub everywhere — `npub1xxxx…yyyy`, the human-readable
spelling of their public key (first 8 + last 4 characters of the full
npub) — across messages and mentions, reactions, huddles, member and
participant lists, the sidebar and channel activity, search, projects,
tray, notifications, and workflow surfaces.

- **Mentions and messages**: key-only mention chips render the compact
npub. Pasting a copied mention back still re-binds it byte-exactly to
the identity it declares, for both the new npub chips and legacy
hex-truncated chips copied by older clients — wrong, missing, or
tampered key qualification is rejected instead of silently degrading to
plain text.
- **Reactions and huddles**: huddle reaction events and the huddle
roster/participants render the compact npub for unnamed participants;
workflow reaction triggers describe authors with the same form.
- **Members and sidebar**: channel and community member lists,
add-member results and invites, the members sidebar, the
channel-activity popover, search, projects (assignees/reviewers/PR
panels), the tray menu, and desktop notifications all fall back to the
compact npub; titles and aria labels keep the machine-readable full
labels.
- **Profile labels**: panel/popover display names and owner handles fall
back to the compact npub (never raw hex) when there is no name;
linked-event (nevent) message metadata shows the npub-shaped author
fallback while the event lookup and event IDs are unchanged.
- **Workflows**: author-picker secondary labels, step destination keys,
and trigger-author references render compact npubs; event and blob IDs
keep their existing hex compacts (they are not identities).
- **Avatars stay distinct**: fallback avatars for key-only identities
derive initials from the key's tail, so prefixed role labels like
"Participant npub1…" no longer collapse every unnamed participant onto
the same initials; people with names keep their name initials.

Preserved exactly: display names and distinct avatars, internal hex keys
(storage/API forms unchanged), clipboard identity roundtrips, event/blob
ID compaction, private keys (no nsec path is touched), and nevent link
handling.

Scope: this PR changes what identity labels **display**, not identity
controls — profile/settings copy controls, the respond-to allowlist,
workflow key fields, and agent dialogs are the sibling slice #7489, and
the shared primitives (`canonicalNpub`, `truncateNpub`, the `<PubKey>`
gate, strict input parsing) come from the foundation #7488.

### Related issue

- Fixes: N/A. Searched existing issues/PRs for duplicates — none found;
the related work is the npub identity stack this slice belongs to.
- Base/dependency: stacks on #7488 (foundation) — this PR does not stand
alone on main.
- #7489 is a sibling slice on the same #7488 base
(profile/agent/workflow controls), not a dependency: this PR does not
require #7489, and #7489 does not require this PR — both only require
#7488.

### Testing

At exact head `4763cbeae1dd521309755e6d61f657324cb98667` (base:
`fix/desktop-npub-identity-d1a` @
`5f3a4a8111998c8aa41ad77cf66992bd1c85343c`; 71 files, +656/−189 —
production +277/−136, test support +379/−53):

- At this head: targeted `mentions.spec.ts` (1/1), the e2e build,
typecheck, and biome — green.
- 9 changed/related unit files: 100/100 green; typecheck, e2e build,
biome, and px text/truncation checks clean; huddle-roster focused run
green; channel-activity e2e 11/11; mutation checks confirm the fallback
wiring (removing it collapses shared initials and drops fallback rows).
- Known pre-existing local e2e failures, unchanged by this PR and
reproduced identically at the upstream merge-base: huddle-transcription
voice-menu attribution (25 pass / 1 fail) and the
`workflow-local-controls` 438px caret drift. Not claimed green locally.
- Update at head `236af9e6137386737e84d3a474d6bc808a704c50` (test-only
follow-ups `1143af345` + `236af9e6`): the `workflow-local-controls`
races were fixed in the test drivers, and the 438px diff was shown to be
a stale Darwin snapshot baseline (name-row enable switch already absent
and `message_posted` already MessageSquare at recording commit
`9390e11c9`) and refreshed — the focused screenshot test, including
keyboard/caret assertions, now passes locally (twice). The full spec was
not rerun after the snapshot refresh; the huddle-transcription item
above is unchanged.

Label/copy text changes are asserted by the e2e specs (`mentions`,
`mention-recipients`, `pubkey-display-screenshots`,
`huddle-transcription`, `channel-activity-popover`,
`workflow-local-controls`) rather than new screenshots; the screenshot
spec pins the compact npub text forms.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): npub identity controls across profile, agents, and workflows (#7489)

🤖
## Summary

Building on #7488's npub foundation, this PR finishes the identity
display change for the controls where you actually manage people and
keys: profile, settings, agent access, and workflows. Everywhere in
these surfaces, an identity key shows — and copies — as its canonical
npub (npub is the human-readable encoding of a Nostr public key: the
compact `npub1j57...fjmv` form where space is tight, the full npub where
the whole key matters), and accepts npub as input.

After this change:

- Profile panel: the public-key row and the managed-by / declared-owner
copies show the full npub. If a key can't be encoded, you see
"Unavailable" with no copy button — never a raw or partial key.
- Settings: the identity card shows and copies the npub. The
hosted-communities account identity derives from the bound key
(`pubkey_hex`) — the same authority as the mismatch gate and hosted
operations — so the display can never disagree with what the app acts
on; an unusable hex falls back to a neutral label instead of rendering
the unverified server npub. The connected claim and a community's
Connect action require that same usable bound key to match the local one
— with no usable binding the card cannot claim connected or start
Connect, while the community list, linking, and delete/rebind recovery
stay available.
- Hosted create/onboarding: the account and device identity rows in the
create flow and owner onboarding derive from the same authoritative
fields (bound key / local key), with the same neutral fallback;
readiness requires a usable bound key that matches the local one.
- Respond-to allowlist (controls who may respond to an agent): entries
can be typed or pasted as hex or npub; both spellings of the same key
are recognized as one entry and dedupe. Search results, chips, and
remove buttons use the compact npub.
- Workflow key fields: to/from keys display as npubs in the form and
save back as canonical hex. Templates like `{{trigger.author}}`, roles,
and free text pass through untouched; placeholders accept both
spellings.
- Recipient and agent dialogs: the verify popover is npub-only (the
raw-hex line is gone); denied-membership screens never show a raw key.
- The Rust-side truncated display name (used for native surfaces) shows
the same compact npub, so those surfaces match the web UI.

Internal representation is unchanged: keys are still stored, sent, and
matched as canonical 64-character hex — npub is a display and input
spelling, normalized to hex at the boundary, so existing data and
integrations keep working. Bound-key usability and comparison use one
normalized form (trimmed, lowercased, 64 hex characters; npub rejected),
so padded or mixed-case spellings of the same key match. Display names,
private keys, and event IDs are untouched.

## Details

- `respondToAllowlist` / `RespondToField`: npub entries normalize to
canonical hex; cross-form dedupe; compact npub in rows and chips;
direct-add accepts npub and stores canonical hex.
- `workflowFormTypes` / `WorkflowStepCard`: hex → npub for display, npub
→ canonical hex on save; templates, roles, and free text pass through in
both directions (roundtrip-tested).
- `UserProfilePanelFields`, `ProfileSettingsCard`,
`HostedCommunitiesSettingsCard`, `MembershipDenied`,
`SelectedRecipientChip`, `AddAgentToChannelDialog`: npub display and
copy; invalid keys → "Unavailable" with no copy; hosted identity rows
derive from the bound `pubkey_hex` (create/onboarding rows from the
bound and local keys), never the unverified server npub;
connected/readiness/Connect gates use the same usable-bound-key
predicate, and the settings Connect invocation callback re-checks it
before starting.
- `src-tauri/src/commands/identity.rs`: `truncated_display_name`
compacts to the first 8 + last 4 characters of the npub (above a 12-char
threshold), mirroring `truncateNpub`.
- e2e: profile key rows and clipboard polls assert npub forms and
raw-hex suppression; the display-screenshots spec pins the npub-only
popover; hosted specs drive the real settings card, create flow, and
onboarding rows through their real providers, and the unlinked/npub-only
identity cases assert no connected claim and no Connect action.

### Related issue

- Fixes: N/A. No separate issue; the related work is the stack below.
- Stack: builds on #7488 (shared npub foundation), now merged; this PR
is rebased onto main and stands on its own.

### Testing

At head `303c90ffa` (base: main `bfc38485`; 24 files, +1125/−146):

- Focused unit suites (respondToAllowlist, workflowFormTypes,
hostedCommunityApi bound-key helpers) green; mutation-checked — dropping
allowlist canonicalization fails the dedupe case, and dropping bound-key
normalization fails the npub-in-hex and padded same-key cases.
- Full desktop unit suite 6,477/6,477, `desktop-typecheck`,
`desktop-check` (formatting fixed narrowly with `biome check --write` on
the touched files only), and a fresh E2E build at the current head; the
add-community + hosted-communities-settings specs 18/18 and onboarding
integration 69/69 on a fresh dedicated port, with focused new-case runs
4+4 covering padded same-key (ready, Connect kept — no false rebind) and
npub-in-hex (neutral label, recovery, no Connect) across the settings
card, create flow, and first-community onboarding, plus the
unlinked-account settings regression asserting Connect cannot occur.
- `cargo fmt`/clippy (both feature sets) and `cargo test identity` (71
pass) passed at the earlier full-change head; since then, the only
production changes in this PR's delta are the hosted identity display
authority and its fail-closed bound-key gating/normalization above
(base-side fixes carry #7488's receipts) — every other change is
test-only.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(mobile): standardize public-key identity display on npub (#7493)

🤖

## Summary

In the mobile app, anyone who hasn't set a display name shows up as a
raw 64-character hex key (e.g. `3a5d4f9c…`) — unreadable, and
unrecognizable as the same identity across screens. Profile and Settings
also let you copy that raw hex. Nostr public keys have a standard
readable form — `npub1…`, the same encoding other Nostr apps and our
desktop app already display. This PR makes every mobile identity surface
render npub instead:

- **Unnamed people everywhere** — message and thread authors, reactions,
typing indicators, member lists, channel details, DM headers and tiles,
inbox, search, forum cards, Pulse notes and reply context, mention
suggestions, and invite rows — now show a compact npub label: first 8 +
last 4 characters of the full npub joined by an ellipsis
(`npub1abcd…wxyz`), the same truncation desktop uses. Previously these
showed truncated raw hex.
- **DM fallback avatars and blank names** — 1:1 DM tiles and headers key
their fallback avatar to the same non-self counterpart the label names,
including self-first participant order; a self-DM keeps its
hex-key-derived initial. Blank or whitespace-only display names fall
back to the compact npub instead of rendering empty, while nonblank
authored names render verbatim (padding included).
- **Profile sheet → "Copy public key"** now copies the full canonical
npub — never raw hex. When the identity string isn't a valid public key,
the copy tile is disabled, so a malformed key never reaches the
clipboard.
- **Settings → Identity (pubkey)** displays and copies the full npub; an
invalid identity reads "Identity unavailable" with copy disabled.
- **Invalid identities never leak truncated raw hex** into the UI
anywhere — they render a neutral "Unknown identity" label.
- **Unchanged on purpose:** display names and verified handles (NIP-05 —
the `name@domain` badge) still render as before. Unnamed avatars keep
distinct per-key initials, derived from the underlying hex key rather
than the npub — otherwise every unnamed key would render the same "N"
initial. Event IDs are not public keys, so they keep their hex
truncation (in Pulse's "Replying to", the parent author shows npub while
an event-id fallback still shows hex). The nevent share link, private
keys, and internal hex storage are untouched. Inputs that accept a key
(invite/member entry) accept both hex and npub and keep working in hex
internally.

### Related issue

N/A. Searched open issues/PRs for npub identity display on mobile —
closest related: none found. Desktop's parallel npub standardization
lives in the stacked desktop PRs (#7488 foundation, #7489 controls,
#7495 display surfaces); this is the independent mobile slice (based
directly on `main`, not on those branches).

### Testing

At exact head `5a620e420a1fd57d9d8011ac26434eed32fcf765` (base: `main`
`44316ff72`; 40 files, +1,345/−154):

- Full mobile suite: 2,098 tests passing (`cd mobile && flutter test`);
`flutter analyze` clean; `dart format --set-exit-if-changed .` clean —
the same checks CI runs.
- Widget/unit coverage at production seams: compact labels and hex-keyed
avatar initials for DM headers/tiles, member rows, mention suggestions,
and Pulse reply context; DM fallback avatars keyed to the labeled
counterpart (self-first order and self-DMs); blank/whitespace
display-name npub fallback with nonblank authored labels verbatim,
including the Activity inbox sender and profile-sheet heading (each with
its own empty/whitespace production-seam regression); full-npub copy and
disabled-copy semantics in profile and settings; invalid-key
suppression; and hex↔npub input round-trips.

Verified via unit and widget tests — no device/simulator validation is
claimed.

### Task provenance

Buzz channel: `1f0e4a3d-7e01-4efe-bb16-843b357f85c9`

Task:
buzz://message?channel=1f0e4a3d-7e01-4efe-bb16-843b357f85c9&id=86b34eb4bd84a1472419e9af22636c011c0fe273e3c196f967d7a36996e149b6

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): order unnamed roster members by full canonical npub (#7503)

🤖

## Summary

- Channel members appear in the Members sidebar. A member who has never
set a display name is listed under an abbreviated form of their public
key (npub), and the sidebar previously sorted those unnamed members by
that short label. Short labels are not unique — different keys can share
one — so the order of unnamed members could look arbitrary or unstable.
Unnamed members now sort by their full public key, so the order is
deterministic.
- When two members display the same name, the previous tiebreak was
membership order (who joined first), which is not visible to a reader
and can shift as roster data loads in. The tiebreak is now the full
public key, so identical display names always land in the same order.
- Nothing gets noisier on screen: the full key is used only for sorting,
and the sidebar still shows the compact abbreviated form. Priorities are
unchanged — authored (custom) names still outrank fallback labels, and
role/current-user grouping still applies.
- Scope is the desktop app's Members sidebar and member management: the
two existing sort comparators. Mobile and other lists in the app are
untouched.

### Related issue

Based on #7495 (introduced the abbreviated npub labels this follows up
on). The original five presentation PRs remain independently reviewable.
No closer duplicate found.

### Testing

- 6469 desktop unit tests, typecheck, and check pass.
- The 3 existing consumer-seam E2E tests still pass; a new E2E test
asserts the sidebar lists unnamed members in full-key order, with
fixture members deliberately inserted in the opposite order so incoming
membership order cannot mask the sort.
- Negative check: reverting only this change makes the new ordering
assertion fail, so it genuinely binds the new sort.
- CI has not run on this PR yet.

Buzz provenance: channel 1f0e4a3d-7e01-4efe-bb16-843b357f85c9 / task
340c3de9b27dbedb8453c0c7652220f9080d30fcc70a7c4f6e27fdd4fa378056

---------

Signed-off-by: Logan Johnson <[email protected]>
Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>

* fix(desktop): require a Codex adapter with Astra support (#7427)

## Summary

Buzz considers codex-acp 1.6.2 current because the supported adapter
floor is still 1.1.7. That adapter bundles Codex 0.148.0, so updating a
separate Codex CLI to 0.153.4 leaves managed agents on the older runtime
and unable to use GPT-6 Astra.

Raise the supported adapter floor to the published 1.10.0 release, which
depends on `@openai/codex ^0.153.3`. Existing discovery and installation
code then classifies older adapters as outdated and offers the managed
reinstall path. Update the availability and install-plan regressions to
cover the observed 1.6.2 installation and the new minimum.

This follows the existing version-floor policy. It does not
automatically update a running installation: the user must complete
Buzz’s offered adapter upgrade. Future upstream compatibility changes
may require another floor update.

### Related issue

No exact duplicate found in searches for Astra, CODEX_PATH, bundled
Codex, outdated runtime, and codex-acp 1.10. Related: #3097 raised the
older floor to 1.1.7 (already present on main); #2422 covers lost error
details for runtime mismatches. Neither resolves this version gap.

Originating conversation:
buzz://message?channel=3286cd76-f83e-4c7d-8317-10a16580744d&id=8b79a73078217222b870fff144c27e7d27bcd5a67c966869c18fe726db716898

### Testing

- Isolated npm install of codex-acp 1.10.0 resolved bundled Codex
0.153.4, with no CODEX_PATH override.
- Live macOS ACP probe: initialize protocol v1 → session/new → select
gpt-6-astra[medium] → prompt. Received `OK` and `stopReason: end_turn`;
usage metadata confirms gpt-6-astra.
- Existing adapter 1.6.2 initialized but advertised no Astra model in
the same probe.
- Desktop Rust formatting and `git diff --check` pass.
- `just desktop-tauri-test`: 3,266 passed, 20 ignored, zero failures
across the Desktop workspace and integration tests.
- Workspace and Desktop Clippy, frontend static checks, and `just
file-size-check` pass.
- Repository `just ci`: still running the remaining
mobile/build/workspace-test stages.

The installed Buzz app and managed adapter were not replaced or
restarted. The live check validates the new adapter/runtime path; a
complete packaged Desktop upgrade workflow remains untested.

Signed-off-by: Stephen DeLorme <[email protected]>

* fix(buzz-acp): report missing models without retrying (#7538)

## Summary

When an agent reports model-not-found, Buzz retries the unavailable
model and delays the failure reply until retries are exhausted. Stop
retrying this error and immediately post a threaded recovery notice. The
notice tells users to select a different model in agent settings, save,
restart the agent to apply the configuration, and re-send their request.

This adds one error-handling branch and regression coverage in
`buzz-acp`. It matches `-32002` errors containing `model not found`.
Other resource-not-found errors, such as stale sessions, retain the
existing retry behavior. Detailed error events remain available for
diagnosis. The existing restart policy is unchanged.

### Related issue

None found in existing issue/PR searches for model-not-found recovery.

### Testing

Playwright captured and visually checked the thread UI with seeded
conversation data and the exact recovery text. The check opens the
request's thread, confirms no reply before the failure, injects the
notice, and verifies the full text is visible. [Before/after
screenshots](https://github.com/block/buzz/pull/7538#issuecomment-5608196506)
show the corrected save-and-restart instructions. These are local test
captures, not a deployed provider recovery flow.

Generated with Codex

---------

Signed-off-by: Diem Nguyen <[email protected]>

* fix(desktop): let inbox title and message author names truncate under narrow panes (#7550)

## Summary

Fixes two instances of the same dead-truncate pattern in the desktop
app, where a flex item's implicit `min-width: auto` prevented `truncate`
from engaging, so long text painted over adjacent controls instead of
ellipsizing:

- **Inbox detail title** (`InboxDetailPane.tsx`): the clickable
context-title button sized to its text instead of shrinking with the
pane, overlapping the header controls (open-in-channel, members, huddle,
more menu). Fixed by adding `max-w-full`.
- **Message author names** (`MessageHeader.tsx` /
`UserProfilePopover.tsx`): the `UserProfilePopover` inline-flex trigger
wrapper refused to shrink below the name's nowrap width, running long
author names under the hover action bar and off the pane edge. Fixed by
adding a `triggerClassName` prop to `UserProfilePopover` and passing
`min-w-0 max-w-full` at the author call site.

Two other suspected instances (project file breadcrumb, drafts pane
title) were stress-tested and already truncate correctly — no change.

### Related issue

N/A — none found.

### Testing

- New Playwright regression tests for both fixes
(`inbox-title-overlap.spec.ts`, `message-author-overlap.spec.ts`,
registered in the smoke project), each proven to discriminate: they fail
with the fix reverted (real measured overlap) and assert the ellipsis
actually engages with non-zero title width, so they can't pass
vacuously.
- Typecheck, lint, and full desktop unit suite green (pre-push hooks);
full desktop e2e smoke suite run earlier: 1402 passed, 3 pre-existing
unrelated failures (each fails identically with the fix reverted).

**Inbox title — before** (long title paints under the header controls):

![Inbox title before: title text overlaps the header control
icons](https://github.com/user-attachments/assets/d5a10f98-114f-4466-8012-468616b82807)

**Inbox title — after** (truncates with ellipsis, controls stay clear):

![Inbox title after: title truncates with an ellipsis before the
controls](https://github.com/user-attachments/assets/80520f7c-1120-4dbc-929d-1ef5eceb874b)

**Author name — before** (long name runs past the header row edge):

![Author name before: name glyphs bleed past the action
bar](https://github.com/user-attachments/assets/be9743ae-ef51-4db0-8c46-a0649e98f0d5)

**Author name — after** (clean cutoff):

![Author name after: name truncates cleanly inside the header
row](https://github.com/user-attachments/assets/24cc0412-ca99-40c4-add6-13380b3ae065)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Signed-off-by: cynfria <[email protected]>
Signed-off-by: Tree Trunks <6ba22921d9dc2ad0aa6ecdf63787ddd24726e266d866da31af69f2e4e146ace5@buzz.block.builderlab.xyz>
Co-authored-by: Claude Fable 5 <[email protected]>
Co-authored-by: Tree Trunks <6ba22921d9dc2ad0aa6ecdf63787ddd24726e266d866da31af69f2e4e146ace5@buzz.block.builderlab.xyz>

* fix(relay): reject presence updates when Redis storage fails (#7532)

## Summary
- Reject kind:20001 presence events with `OK false` / `error: presence
storage unavailable` when Redis SET or DEL fails, before publishing,
local fan-out, or local-event marking.
- Preserve the producer contract needed by snapshot-confirming
consumers: delivered live presence must follow successful mutation of
the Redis state read by snapshots.
- Classify those backend rejections with the existing `IngestError`
taxonomy so a presence storage outage counts as
`buzz_events_rejected_total{transport="ws",reason="error"}`, not client
`reason="invalid"`; genuine client-input refusals (verification failure,
membership gates) stay `invalid`, and every wire message is an unchanged
fixed sanitized string (review follow-up, no protocol wording change).
- Add actual `handle_event` integration coverage for rejected
online/offline transitions, healthy online→offline
accepted/stored/fanned-out behavior, and the rejection-counter routing
on storage failure with an invalid-signature control.

This is standalone on main; it does not depend on the mobile
implementation. Deploy this relay prerequisite before relying on #7526's
snapshot-confirmation policy. Existing
pubsub-failure-after-successful-storage behavior and disconnect TTL
cleanup are deliberately unchanged. A storage error may be an ambiguous
write outcome, not a rollback guarantee; the rejected event is not
published by this handler. Clients may retry the generic `error:`
rejection. Desktop's 60s heartbeat retries non-offline presence, not
every explicit offline transition.

### Related issue
Addresses the relay prerequisite identified in [#7526 review
5157607827](https://github.com/block/buzz/pull/7526#pullrequestreview-5157607827).
Searched open presence/storage PRs; no duplicate relay storage-error
rejection fix found. #7382/#7383/#7526 heads and bases are unchanged.

### Testing
Exact head: `389174df29cc02d0f885c03209eff661d8bb2ec0` (+380/-13; 393
total), one commit `389174df2` on top of the reviewed `c031d6eb1`
(DCO-signed; base `bfc384855889432df4a333a0edf3080f332ee169` unchanged).

- PASS: `cargo fmt --all -- --check`, `cargo clippy -p buzz-relay
--all-targets -- -D warnings`, `git diff --check`, `just
file-size-check`, PostgreSQL discovery validation — all run at the exact
final head with a clean tree before and after.
- PASS: documented native `scripts/postgres-test-run.sh -p buzz-relay
--lib --tests`: **89/89** actual integration tests, including the four
presence cases (online/offline storage rejection, healthy
online→offline, and the new rejection-classification case). Owned
PostgreSQL 17/Redis on isolated loopback ports, schema plus
reconciliation applied; no shared development database.
- PASS: explicit `cargo test -p buzz-relay presence_storage -- --ignored
--nocapture`: **4/4**, not skipped.
- Full isolated relay crate suite at the final head (`cargo nextest run
-p buzz-relay --lib --tests`): **1062 run: 1062 passed, 94 skipped**.
The previously failing
`api::mesh_demo::tests::demo_join_forwarded_arm_round_trips_echo` passed
in this run (1.5s); it is a known timing-sensitive main baseline failure
tracked open in #7140 and untouched by this PR, so this single passing
run is reported as-is and does not claim environmental clearance or
close #7140. No full-suite-green claim is made beyond this run.
- Mobile is untouched; #7526's existing 2090-test/format/analyze
evidence remains scoped to its unchanged head. Its separate Desktop
Smoke E2E (2) failure remains red; no CI retries requested.

[Production-seam regression
coverage](https://github.com/block/buzz/blob/389174df29cc02d0f885c03209eff661d8bb2ec0/crates/buzz-relay/src/handlers/event.rs#L1491-L1803):
the metric case drives real `handle_event` traffic against a genuinely
dead Redis endpoint with a seeded active PostgreSQL community and a
registered presence watcher, asserts the storage rejection counts
`reason="error"` while a tampered-signature control through the same
dispatcher arm stays `reason="invalid"`, and re-asserts the rejected
ACK, no fan-out, and no local-event marker. Counter assertions use a
thread-local recorder guard held across `.await` points (the buzz-db
counter-test convention) inside the per-process nextest postgres-ci
lane, so no parallel test can race the counter snapshot.

No UI change or screenshot. Local logs and reproducible service/gate
scripts are retained under
`WORK_LOGS/MOBILE_FEEDBACK_PRESENCE_20260909/relay_prerequisite/metric_correction/`
in the engineering workspace. This PR is a review candidate, not merge
clearance.

Causal checks: restoring only the pre-fix production mutation block
makes both original rejection tests fail (`OK true` instead of `false`);
healthy success still passes. Reverting only the typed classification
(mapping the ephemeral `Internal` arm back to `invalid`) makes the new
metric regression fail with the outage counted as `[("ws","invalid",2)]`
instead of `[("ws","error",1),("ws","invalid",1)]`. The unchanged mesh
echo case also failed 504/200 with the main-production block restored in
the prior run, supporting its separation from this change without
claiming environmental clearance. Candidate source restored
byte-for-byte after each mutation. Repository-wide `just ci` was not
rerun; the scoped relay gates above are the new evidence.

---------

Signed-off-by: Logan Johnson <[email protected]>

* fix(markdown): align mention chip wrapping (#7501)

**Category:** fix
**User Impact:** Human and agent mentions now break across lines with
the same cloned chip treatment as repository and permalink chips while
preserving the conversation text rhythm.

**Problem:** Profile-backed rendered mentions sat inside an
`inline-flex` popover trigger, unlike entity chips, so the wrapper
interfered with true inline fragmentation. The browser-layout test
measured text-range rows rather than the painted chip rectangles,
allowing touching decorations to pass as “separate” fragments.

**Solution:** Keep the profile trigger interactive but override its
layout to true `inline`, then give mention fragments 18px computed
leading inside the message’s 20px prose rhythm. Chromium paints each
fragment at 17px and advances it by 20px, leaving a visible gap between
cloned rounded rectangles. The browser test now measures the chip’s own
`getClientRects()` and asserts fragment count, height, gap, and step;
entity links retain their existing 22px leading.

<details>
<summary>File changes</summary>

**desktop/src/features/profile/ui/UserProfilePopover.tsx**
Allows inline consumers to override the trigger wrapper’s layout without
changing other profile-popover call sites.

**desktop/src/shared/styles/globals/markdown.css**
Keeps one shared wrapping-chip mechanic and gives mention decorations
enough room to separate visibly within 20px prose.

**desktop/src/shared/ui/markdown.test.mjs**
Pins both rendered mentions and entity links to the shared wrapping-chip
contract.

**desktop/src/shared/ui/markdown/MarkdownMention.tsx**
Makes the profile-popover trigger truly inline so the nested mention
chip can fragment with surrounding prose.

**desktop/src/shared/ui/mentionChip.ts**
Keeps `wrapping-inline-chip` as the single contract for fragmenting
decorated chips.

**desktop/tests/e2e/mentions.spec.ts**
Measures the painted chip rectangles, requires a positive fragment gap,
and verifies the inline trigger remains mouse- and keyboard-operable.

**desktop/tests/e2e/navigation.spec.ts**
Keeps a wrapped repository chip as the control, asserting its existing
22px line height and fragment advance.

</details>

## Reproduction steps

1. Open a channel in Buzz Desktop using dark theme.
2. Send a message containing a human mention and another containing an
agent mention; both chips should remain aligned with adjacent text on a
20px line.
3. Render a collision-qualified mention in a narrow message width; it
should break into separately decorated fragments exactly like another
wrapping chip, while each fragment follows the 20px prose rhythm.
4. Render a long repository or permalink chip in the same constrained
width; it should retain its roomier 22px fragment spacing.

## Screenshot

The dark-theme production renderer shows the real qualified label (`bob
(npub1hv3…tpuc)`) at an 8rem width. The two lines now paint as visibly
separate rounded fragments rather than one continuous rectangle.

![Qualified human mention wrapping into two visibly separate rounded
fragments in the dark-theme Buzz
timeline](https://github.com/user-attachments/assets/ee6e7be5-937e-4022-9c82-cf71f8203470)

## Validation

At commit `2b063e1b4ade30e11f1616269ad4ba4190366885`:

- Pre-push desktop gates — file-size check, Biome/checks, typecheck, and
6,483 unit tests passed
- `pnpm --dir desktop build` — passed
- Focused Playwright coverage for single-line agent mention, single-line
human mention, wrapped qualified mention including keyboard profile
activation, and timeline mention click — 4 passed
- `git diff --check` — passed

---------

Signed-off-by: Taylor Ho <[email protected]>
Co-authored-by: Rizz <[email protected]>
Co-authored-by: Carl <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>

* fix(acp): integrate the Buzz Pi adapter fork (#7552)

## Summary

PR #7335 worked around missing Pi adapter support by generating a
private Pi launcher and injecting Buzz's standing prompt and skills at
process launch. The Buzz Pi fork now carries the required adapter
extensions, so this removes that launcher and returns prompt
construction to the normal ACP session path while retaining the
base-prompt composition introduced by #7335.

The Pi preset now installs `salman1993/pi-acp` and launches its renamed
`buzz-pi-acp` binary. Buzz adds `-- --skill
<harness-cwd>/.agents/skills` only when launching that binary, sends the
complete composed prompt as the `_meta.systemPrompt` replacement string
on `session/new` only when `initialize.agentInfo.name` is `buzz-pi-acp`,
and sends the scoped title alongside it as `_meta.sessionTitle`. The
fork identity is treated as system-prompt capable regardless of its
reported ACP protocol version, which prevents duplicate legacy
user-message framing. Upstream `pi-acp` does not receive either
fork-specific behavior. Observer transcript projection accepts the
string, `{ replace }`, and `{ append }` metadata forms.

The fork now stores restore metadata in one atomic file per session
under `~/.pi/buzz-pi-acp/sessions/`. This prevents concurrent Buzz
workers from overwriting another session's prompt or title. The fix
landed in
[salman1993/pi-acp#9](https://github.com/salman1993/pi-acp/pull/9).

This supersedes the closed #7508. No agent-configuration rules changed;
this changes the Buzz Pi adapter contract and launch arguments.

### Related issue

#7329

### Testing

Installed fork commit `09cf07e436b8f18e52401558f988f31a15702313` through
the documented Git URL. The installed bundle matched the committed
`dist/index.js` byte for byte and contained the `~/.pi/buzz-pi-acp`
metadata path. The fork's 106 non-skipped tests, typecheck, and lint
pass.

Ran the ignored real-Pi integration test through Buzz's production
session composer. The test exercised the renamed package,
`agentInfo.name`, and the new per-session metadata store. Base, persona,
team, core-memory, huddle, canvas, and skill markers each appeared once
after switching sessions and again after restarting the adapter, while
the other session and Pi's native default prompt were absent.

Added regression coverage proving `buzz-pi-acp` receives fork-specific
system-prompt metadata and managed skills while upstream `pi-acp` does
not. `just ci` passes.

Generated with Codex

---------

Signed-off-by: Salman Mohammed <[email protected]>

* feat(git): add default-branch management to relay and CLI (#7562)

Authored by Brain and opened on behalf of Wes (`wesbillman`).

##…