Co-authored-by: Anthony Eid <hello@anthonyeid.me>

Co-authored-by: Anthony Eid <hello@anthonyeid.me>

Co-authored-by: Anthony Eid <hello@anthonyeid.me>

Harden the terminal tool's permission system to reject commands
containing shell substitutions and interpolations (`$VAR`, `${VAR}`,
`$(…)`, backticks, `$((…))`, `<(…)`, `>(…)`) before they reach terminal
creation.

## Changes

### Shell command parser (`shell_command_parser`)
- Added structured terminal command-prefix extraction with env-var
prefix support
- Added parser-backed validation that classifies commands as
Safe/Unsafe/Unknown
- Extended normalized command extraction to include scalar env-var
assignments in order
- Preserved quoted assignment values when they contain whitespace or
special characters

### Pattern extraction (`agent/pattern_extraction`)
- Updated terminal pattern extraction to use structured parser output
- Included env-var prefixes in generated allow patterns
- Normalized regex token boundaries to `\s+` while preserving display
whitespace

### Tool permissions (`agent/tool_permissions`)
- Added invalid-terminal-command rejection for forbidden
substitutions/interpolations
- Added unconditional allow-all bypass (global default Allow, or
terminal-specific Allow with empty patterns)
- Preserved hardcoded denial precedence over allow-all

### Terminal tool (`agent/tools/terminal_tool`)
- Updated tool description and input schema to explicitly prohibit shell
substitutions
- Added comprehensive SEC-264 regression test suite (20 new tests)
covering:
- All forbidden constructs (`${HOME}`, `$1`, `$?`, `$$`, `$@`,
`$(whoami)`, backticks, `$((1+1))`, `<(ls)`, `>(cat)`, env-prefix
variants, multiline, nested)
  - Allow-all exception paths (global and terminal-specific)
  - Hardcoded-denial precedence
- Env-prefix permission flow (matching, value mismatch rejection,
multiple assignments, quoted whitespace)

Closes SEC-264

Release Notes:

- Terminal tool permissions regexes can now match environment variables
(e.g. `FOO=bar cmd arg1 arg2`)
- If terminal tool permissions have active permissions regexes running
on them, then bare interpolations like `$FOO` are disallowed for
security, since regexes wouldn't be able to match on them.

Reverts https://github.com/zed-industries/zed/pull/51241

Release Notes:

- N/A

Closes #48393

When switching the language of an unsaved buffer (e.g., from Plain Text
to JavaScript), the editor's edit_prediction_settings were not
recalculated. This caused the "This Buffer" toggle in the edit
prediction menu to show a stale state, making the first click a no-op. I
added a call to update_edit_prediction_settings in the LanguageChanged
event handler so the editor immediately reflects the correct edit
prediction state for the new language.

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Fixed the "Show Edit Predictions For This Buffer" toggle being a no-op
in unsaved buffers after switching the buffer's language.

Closes #50785

Opening a .wav file (PCM 16-bit) caused Zed to freeze because the binary
detection heuristic in `analyze_byte_contentmisidentified` it as
UTF-16LE text. The heuristic determines UTF-16 encoding solely by
checking whether null bytes are skewed toward even or odd positions. PCM
16-bit audio with small sample values produces bytes like `[sample,
0x00]`, creating an alternating null pattern at odd positions that is
indistinguishable from BOM-less UTF-16LE by position alone.

### Why not just add more binary headers?

The initial approach
(https://github.com/zed-industries/zed/commit/32d8bd700943ae7cb46231bc08d752d8b7987432)
was to add audio format signatures (RIFF, OGG, FLAC, MP3) to known
binary header. While this solved the reported `.wav` case, any binary
format containing small 16-bit values (audio, images, or arbitrary data)
would still be misclassified. Adding headers is an endless game that
cannot cover unknown or uncommon formats.

### Changes

* Adds `is_plausible_utf16_text` as a secondary validation: when the
null byte skew suggests UTF-16, decode the bytes and count code units
that fall in C0/C1 control character ranges (U+0000–U+001F,
U+007F–U+009F, excluding common whitespace) or form unpaired surrogates.
Real UTF-16 text has near-zero such characters. I've set the threshold
at 2% — note that this is an empirically derived value, not based on any
formal standard.

**Before fix**

<img width="1147" height="807" alt="스크린샷 2026-03-06 오후 9 00 07"
src="https://github.com/user-attachments/assets/2e6e47f9-f5e7-4cab-9d41-cc3dd20f9142"
/>

**After fix**
<img width="1280" height="783" alt="스크린샷 2026-03-06 오전 1 17 43"
src="https://github.com/user-attachments/assets/3fecea75-f061-4757-9972-220a34380d67"
/>


Before you mark this PR as ready for review, make sure that you have:
- [X] Added a solid test coverage and/or screenshots from doing manual
testing
- [ ] Done a self-review taking into account security and performance
aspects
- [ ] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Fixed binary files (e.g. WAV) being misdetected as UTF-16 text,
causing Zed to freeze.

Closes #34976

## Root Cause

The current implementation of `fmt::Display` in
`KeyBindingContextPredicate` hardcodes a parentheses around `And` and
`Or` cases. Causing the text displayed on the keymap editor to have
nested parentheses.

This PR refactors the display logic and make the parentheses added only
when there's nested operator.

## As is/ To be

As is | To be
--- | ---	
<img width="800" height="1101" alt="Screenshot 2026-03-06 at 15 07 10"
src="https://github.com/user-attachments/assets/33b4366d-4d16-48db-9cc6-b79d6cc7ddc7"
/> | <img width="800" height="977" alt="Screenshot 2026-03-06 at 14 54
57"
src="https://github.com/user-attachments/assets/3d5ed38d-949d-4265-8fad-e29250c3a285"
/>

## Test coverage

<img width="800" height="916" alt="Screenshot 2026-03-06 at 15 00 42"
src="https://github.com/user-attachments/assets/5846b908-5c86-48a7-919e-a7ae198484a8"
/>



Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [ ] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- keymap_editor: Remove redundant parentheses when displaying keymap
context

---------

Co-authored-by: Ben Kunkle <ben@zed.dev>

This matches how the `buck2-lsp` works for that extension.

Fixes https://github.com/zaucy/zed-starlark/issues/18

Before you mark this PR as ready for review, make sure that you have:
- [ ] (n/a) Added a solid test coverage and/or screenshots from doing
manual testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- N/A

This PR bumps the update action again. It also fixes a bug where a cache
issue could occur during the extension auto bump and adds update
auto-merge for PRs from staff-members in zed-industries/extensions.

Release Notes:

- N/A

Now that the migration has happened, we can stop making these ever
present

Release Notes:

- N/A

Release Notes:

- N/A

Dramatically simplifies what "New Worktree" does when forking a
workspace.

Previously we captured the entire dock structure and all open file
paths, then restored them all in the new worktree workspace. Now the
behavior is:

- If you have an active buffer open, we reopen just that one file (no
other buffer state is carried over).
- If you do not have an active buffer open, we instead fullscreen (zoom)
the agent panel (equivalent to cmd-esc).

No other workspace state is carried over.

Implementation details:
- Replace `capture_dock_state` / `set_dock_structure` /
`open_item_abs_paths` with a single `active_file_path` lookup.
- Pass `None` for the `init` closure instead of restoring dock layout.
- Only reopen the single active file via `open_paths`, with proper error
logging (`detach_and_log_err`) instead of silent `.detach()`.
- Emit `PanelEvent::ZoomIn` on the agent panel when no active buffer
exists, after `focus_panel` to ensure the panel is properly activated
first.
- Warn when an active file exists but cannot be remapped to the new
worktree.

Closes AI-87

Release Notes:

- N/A

When the user selected a custom agent (e.g. Codex) and chose "Start
Thread in New Worktree", the new workspace's panel ignored the selection
and defaulted to Zed Agent (NativeAgent).

The fix captures `self.selected_agent()` in
`handle_worktree_creation_requested` before the async task, passes it
through to `setup_new_workspace`, and forwards it to `external_thread`
instead of `None`.

Closes AI-84

(No release notes because it's still feature-flagged.)

Release Notes:

- N/A

## Summary

- Update the context window table in `models.md` to reflect that Claude
Sonnet 4.6 and Opus 4.6 now support 1M token context windows (previously
listed as 200k)
- Remove Sonnet 4.5/4.6 from the "may increase in future releases"
callout since the 4.6 models are now at 1M

Release Notes:

- N/A

We now share connections across multiple "ConnectionView"s the naming
does not make sense anymore

Release Notes:

- N/A

Release Notes:

- N/A *or* Added/Fixed/Improved ...

Closes discussion #34890 

This is similar to the vim prev/next method/section motion, but more
flexible because this follows the items in editor's outline (Tree sitter
or LSP provided).

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Added actions `editor::GoToPreviousSymbol` and
`editor::GoToNextSymbol` actions to go to the previous and next outline
symbol. This is either the tree sitter outline, or the LSP provided
outline depending on the configuration.

We now share connections across multiple "ConnectionView"s the naming
does not make sense anymore

Release Notes:

- N/A

Fixes ZED-5VQ

Release Notes:

- N/A *or* Added/Fixed/Improved ...

The Python (debugpy) and Go (delve-shim-dap) debug adapters
unconditionally require network access on every debug session start —
even when the adapter binary is already cached locally. This makes the
debugger completely unusable offline.

Python: fetch_debugpy_whl → maybe_fetch_new_wheel hits pypi; failure
propagates as a fatal error.
Go: install_shim → fetch_latest_adapter_version hits the GitHub API;
failure propagates as a fatal error.
CodeLLDB and JavaScript adapters already handle this correctly.

Fix for Python and Go:

Wrap the network fetch in each adapter with a fallback: if the request
fails and a previously downloaded adapter exists on disk, log a warning
and use the cached version. If no cache exists, the original error
propagates unchanged.

Logs after the fix:

```
2026-03-06T16:31:51+05:30 WARN  [dap_adapters::python] Failed to fetch latest debugpy, using cached version: getting latest release
2026-03-06T16:31:51+05:30 INFO  [dap::transport] Debug adapter has connected to TCP server 127.0.0.1:45533
```

Limitations: 

The debugger must be run at least once with internet connectivity to
populate the cache on disk. This PR does not change that requirement.


Closes #45781

Before you mark this PR as ready for review, make sure that you have:
- [ ] Added a solid test coverage and/or screenshots from doing manual
testing
- No automated tests are included. The existing MockDelegate stubs
http_client() and fs() as unimplemented!(), so testing the fallback path
would require new mock infrastructure. The fix was verified manually by
running the debug build offline with and without cached adapters. The
CodeLLDB adapter's equivalent fallback (lines 357-374) also has no
dedicated test.
- [x] Done a self-review taking into account security and performance
aspects

Release Notes:

- Fixed debugger failing to start when offline if a debug adapter was
previously downloaded.

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- N/A

---------

Co-authored-by: cameron <cameron.studdstreet@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>

Release Notes:

- Fixed: echo's on experimental audio pipeline

Some more context:
- `EchoCanceller` was not implemented on the output side
- removes the planned migration to a different sample rate (44100) and
channel count (1)
- de-duplicate the interleaved `#cfg[]`'s and centralized them in
`echo_canceller.rs`

This PR bumps the version of the GLSL extension to v0.2.2.

Release Notes:

- N/A

Co-authored-by: zed-zippy[bot] <234243425+zed-zippy[bot]@users.noreply.github.com>

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Updated our BYOK integration to support the new 1M context windows for
Opus and Sonnet.

Mostly just a test to check whether everything works properly now.

Release Notes:

- N/A

This PR adds the missing methods for the `align-self` css property. The
`align_self` field existed on the `StyleRefinement` struct but was
inaccessible using the `styled` trait.

Release Notes:

- Adds `self_start`, `self_end`, `self_center`, `self_flex_start`,
`self_flex_end`, `self_baseline`, and `self_stretch` methods to the
`trait Styled`

## Summary

- Implements `icon_label` on `GitPanel` to return the total count of
uncommitted changes (`new_count + changes_count`) when non-zero, capped
at `"99+"` for large repos.
- Updates `PanelButtons::render()` to render that label as a small green
badge overlaid on the panel's sidebar icon, using absolute positioning
within a `div().relative()` wrapper.
- The badge uses `version_control_added` theme color and
`LabelSize::XSmall` text with `LineHeightStyle::UiLabel` for accurate
vertical centering, positioned at the top-right corner of the icon
button.

The `icon_label` method already existed on the `Panel`/`PanelHandle`
traits with a default `None` impl, and was already implemented by
`NotificationPanel` (unread notification count) and `TerminalPanel`
(open terminal count) — but was never rendered. This wires it up for all
three panels at once.

## Notes

- Badge is positioned with non-negative offsets (`top(0)`, `right(0)`)
to stay within the parent container's bounds. The status bar's
`render_left_tools()` uses `.overflow_x_hidden()`, which in GPUI clips
both axes (the `overflow_mask` returns a full content mask whenever any
axis is non-`Visible`), so negative offsets would be clipped.
- `LineHeightStyle::UiLabel` collapses line height to `relative(1.)` so
flex centering aligns the visual glyph rather than a
taller-than-necessary line box.
- No new data tracking logic — `GitPanel` already maintains `new_count`
and `changes_count` reactively.
- No feature flag or settings added per YAGNI.

## Suggested .rules additions

The following pattern came up repeatedly and would prevent future
sessions from hitting the same issue:

```
## GPUI overflow clipping

`overflow_x_hidden()` (and any single-axis overflow setter) clips **both** axes in GPUI.
The `overflow_mask()` implementation in `style.rs` returns a full `ContentMask` (bounding box)
whenever any axis is non-`Visible`. Absolute-positioned children that extend outside the element
bounds will be clipped even if only the X axis is set to Hidden.
Avoid negative `top`/`right`/`bottom`/`left` offsets on absolute children of containers
that have any overflow hidden — keep badge/overlay elements within the parent's bounds instead.
```

Release Notes:

- Added a numeric badge to the git panel sidebar icon showing the count
of uncommitted changes.

---------

Co-authored-by: Danilo Leal <daniloleal09@gmail.com>

Closes https://github.com/zed-industries/zed/issues/51576

Release Notes:

- Select first entry in the language selector when matching a query



https://github.com/user-attachments/assets/c824c024-d2f1-416e-a347-0eab7bc3ae0a

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>

This should put us in a state where everything finally works in its
entirety.

Release Notes:

- N/A

Closes #43329

## Summary
This fixes `path:line:column` navigation for files containing non-ASCII
text.

Before this change, open path flows were passing the external column
directly into `go_to_singleton_buffer_point`. That happened to work for
ASCII, but it was wrong for Unicode because external columns are
user-visible character positions while the editor buffer stores columns
as UTF-8 byte offsets.

This PR adds a shared text layer conversion for external row/column
coordinates and uses it in the affected open-path flows:
 - file finder navigation
 - recent project remote connection navigation
 - recent project remote server navigation

It also adds regression coverage for the Unicode case that originally
failed.

As a small - necessary - prerequisite, this also adds
`remote_connection` test support to `file_finder`'s dev-deps so the
local regression test can build and run on this branch. That follows the
same feature mismatch pattern previously fixed in #48280. I wasn't able
to locally verify the tests/etc. w/o the addition... so I've rolled it
into this PR. Tests are green.

The earlier attempt in #47093 was headed in the right direction, but it
did not land and did not include the final regression coverage requested
in review.

## Verification
 - `cargo fmt --all -- --check`
 - `./script/clippy -p text`
 - `./script/clippy -p file_finder`
 - `./script/clippy -p recent_projects`
 - `cargo test -p file_finder --lib --no-run`
- `cargo test -p file_finder
file_finder_tests::test_row_column_numbers_query_inside_file -- --exact`
- `cargo test -p file_finder
file_finder_tests::test_row_column_numbers_query_inside_unicode_file --
--exact`
- `cargo test -p text
tests::test_point_for_row_and_column_from_external_source -- --exact`

## Manual

I reproduced locally on my machine (macOS) w/ a stateless launch using a
Unicode file (Cyrillic)

Before:
 - `:1:5` landed too far left
 - `:1:10` landed around the 4th visible Cyrillic character

After:
 - `:1:5` lands after 4 visible characters / before the 5th
 - `:1:10` lands after 9 visible characters / before the 10th

Release Notes:
- Fixed `path:line:column` navigation so non-ASCII columns land on the
correct character.

---------

Co-authored-by: Kirill Bulatov <kirill@zed.dev>

Closes #ISSUE

Before you mark this PR as ready for review, make sure that you have:
- [ ] Added a solid test coverage and/or screenshots from doing manual
testing
- [ ] Done a self-review taking into account security and performance
aspects
- [ ] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- N/A *or* Added/Fixed/Improved ...

Co-authored-by: Oleksiy <oleksiy@zed.dev>

This PR is motivated by internal feedback in which the notification that
we show inviting to resolve merging conflicts with an agent also pops up
if the agent itself ran `git merge`. In this case, the notification is
unnecessary noise. So, what I'm doing here is simply _not_ showing it if
there's a running agent.

I want to note that this change is accepting a trade-off here, in which
there could be cases that even if an agent is running, the notification
can still be useful. There could be other ways to identify whether the
agent is running `git merge`, but they all felt a bit too complex for
the moment. And given this is reasonably an edge case, I'm favoring a
simple approach for now.

Release Notes:

- N/A

---------

Co-authored-by: Lukas Wirth <lukas@zed.dev>

This PR bumps the version of the HTML extension to v0.3.1.

Release Notes:

- N/A

Co-authored-by: zed-zippy[bot] <234243425+zed-zippy[bot]@users.noreply.github.com>

Moving to the for here before the fix gets upstreamed. We might revisit
this at a later point anyway.

Release Notes:

- N/A

Closes https://github.com/zed-industries/zed/discussions/49740

Adds optional file and folder icons to the Git panel so its file list is
easier to scan, especially in larger repositories.

- add folder icons for Git panel entries
- add Git panel settings for file and folder icon visibility

---

Release Notes:

  - Made the Git Panel aware of icon themes.
  - Added the ability to render file type icons in the Git panel.

---------

Co-authored-by: Danilo Leal <daniloleal09@gmail.com>

Reverts zed-industries/zed#51400

## Summary

Closes #46126

Dim terminal colors in **Ayu Dark** and **Ayu Mirage** were defined as
light pastels instead of muted/darker versions of the normal colors. On
a dark background, the APCA minimum contrast algorithm boosted these
already-light colors toward white, making "dimmed" text (e.g.
zsh-autosuggestions) appear brighter than normal text.

### Root cause

| Color | Ayu Dark (before) | Ayu Dark (after) | Pattern |
|-------|-------------------|------------------|---------|
| dim_foreground | `#0d1016` (= background!) | `#85847f` | Muted
foreground |
| dim_red | `#febab9` (light pink) | `#a74f53` | 70% of normal red |
| dim_green | `#d8eca8` (light pastel) | `#769735` | 70% of normal green
|
| dim_yellow | `#ffd9aa` (light pastel) | `#b17d3a` | 70% of normal
yellow |

The fix follows the same convention as **Gruvbox** and **One Dark**,
where dim = ~70% brightness of the normal color.

**Ayu Light** was already correct (dim colors are darker, which is
correct for light backgrounds).

## Test plan

- [ ] Open terminal in Zed with Ayu Dark theme
- [ ] Enable zsh-autosuggestions (or any tool that uses dim/faint ANSI
colors)
- [ ] Type a partial command to trigger autosuggestions
- [ ] Verify suggested text appears as muted/dim, not bright white
- [ ] Repeat with Ayu Mirage theme
- [ ] Verify Ayu Light theme is unaffected

---

Release Notes:

- Improved some Ayu Dark/Mirage theme's terminal colors.

Co-authored-by: Danilo Leal <daniloleal09@gmail.com>

Closes #47900 

## Root cause

The current markdown preview only re-renders on `EditorEvent::Edited,
DirtyChanged, ExcerptsEdited`, but agent edits are implemented via
[`buffer.edit()`](https://github.com/dongdong867/zed/blob/eb3f92708b6dc67bb534c1c44a200c0cb5c4997b/crates/agent/src/edit_agent.rs#L375)
which does not guaranty to emit the `EditorEvent::Edited` event. Causing
the markdown preview stuck on the last received parsed markdown.

## Applied fix

Subscribing to `EditorEvent::BufferEdited` when initializing the
markdown preview view. This will cause the view to update when received
`BufferEdited` event including agent edits to the file.

## As is/ To be

As is | To be
--- | ---
<video
src="https://github.com/user-attachments/assets/a0b13467-7758-4572-ae01-bcbc40beff6c"
/> | <video
src="https://github.com/user-attachments/assets/3b5463f3-46ad-4fe6-8563-ceb4347035db"
/>

---

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [ ] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Fixed markdown preview not re-rendering when edited by agent

#### Fixes #51082

#### Description :

When a user configures `default_config_options` (e.g. a specific model)
for an ACP agent, the setting was only applied when creating a new
session. Resuming or loading an existing session (e.g. from history)
skipped this step, causing the model dropdown to revert to "Default
(recommended)".

#### Fix : 

Extract the config options defaulting logic into a shared helper
apply_default_config_options and call it consistently across all three
session entry points: new_session, resume_session, and load_session
Release Notes:

#### Release notes :
Fixed model dropdown resetting to "Default (recommended)" when opening a
previous conversation from history. Configured models (via
default_config_options) are now correctly applied when resuming or
loading existing sessions.


#### Video :
[Screencast from 2026-03-15
12-55-58.webm](https://github.com/user-attachments/assets/977747b9-390c-4f78-91fc-91feace444e1)

Closes #45135

Before you mark this PR as ready for review, make sure that you have:
- [x] Added a solid test coverage and/or screenshots from doing manual
testing
- [x] Done a self-review taking into account security and performance
aspects
- [x] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Fixed project root name appearing in file paths while searching in
file finder.

Quick-and-dirty version as we're trying this out with the first cohort.

Release Notes:

- N/A

Fixes #29774

When filtering the outline panel, matches with equal fuzzy scores
previously defaulted to the last item due to iterator `max_by_key`
semantics. This caused the bottommost match (e.g. `C::f`) to always be
pre-selected regardless of cursor position.

Changes:
- Select the match nearest to the cursor when scores are tied, using a
multi-criteria comparison: score -> cursor containment depth ->
proximity to cursor -> earlier index
- Move outline search off the UI thread (`smol::block_on` -> async
`cx.spawn_in`) to avoid blocking during filtering
- Wrap `Outline<Anchor>` in `Arc` for cheap cloning into the async task
- Add `match_update_count` to discard results from stale queries

Tests : 
Adds a regression test:
`test_outline_filtered_selection_prefers_cursor_proximity_over_last_tie`
which passes

Video : 
[Screencast from 2026-03-03
17-01-32.webm](https://github.com/user-attachments/assets/7a27eaed-82a0-4990-85af-08c5a781f269)


Release Notes:
Fixed the outline filtering always select last match

---------

Co-authored-by: Kirill Bulatov <kirill@zed.dev>

We had inconsistent handling of this read-only behavior. Now subagents
should behave the same as agents that don't support editing

Release Notes:

- agent_ui: Fix inconsistent behavior for subagent views when focusing
on previous messages.

Co-authored-by: Jakub Konka <kubkon@jakubkonka.com>

Closes #ISSUE

Before you mark this PR as ready for review, make sure that you have:
- [ ] Added a solid test coverage and/or screenshots from doing manual
testing
- [ ] Done a self-review taking into account security and performance
aspects
- [ ] Aligned any UI changes with the [UI
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)

Release Notes:

- Improve recovery of audio latency after CPU-intensive period.

Co-authored-by: Jakub Konka <kubkon@jakubkonka.com>

Fixes ZED-4YP

Sort and deduplicate anchor ranges in do_helix_select before passing
them to select_anchor_ranges. When the search wraps past the end of the
document back to the beginning, the new selection is at a lower offset
than the accumulated prior selections, producing unsorted anchors that
crash the rope cursor with 'cannot summarize backward'.


Release Notes:

- Fixed a panic in helix mode with search selecting wrapping around the
document end

When a command used an explicit register (e.g. `"_dd` or `"add`), the
subsequent dot repeat (`.`) was ignoring that register and using the
default instead.

Store the register at recording start in `recording_register_for_dot`,
persist it to `recorded_register_for_dot` when recording stops, and
restore it in `Vim::repeat` when no explicit register is supplied for
`.`. An explicit register on `.` (e.g. `"b.`) still takes precedence.

This commit also updates the dot-repeat logic to closely follow Neovim's
when using numbered registers, where each dot repeat increments the
register. For example, after using `"1p`, using `.` will repeat the
command using `"2p`, `"3p`, etc.

Closes #49867

Release Notes:

- Fixed vim's repeat . to preserve the register the recorded command
  used
- Updated vim's repeat . to increment the recorded register when using
  numbered registers

---------

Co-authored-by: dino <dinojoaocosta@gmail.com>