Right now agent extensions can specify icon paths that point anywhere.
This changes it so that they can only specify icon paths that are
subdirectories of the extension's root dir.
Release Notes:
- Restrict agent server extension icon paths to subdirectories of the
extension's root directory
The default behavior for Vim search with `*` and `#` in normal mode is
to initiate a search and immediately jump to the next or previous match
respectively.
This behavior can be annoying, so Vim has many plugins to address this
specifically:
- [vim-asterisk](https://github.com/haya14busa/vim-asterisk)
-
[vim-SearchHighlighting](https://github.com/inkarkat/vim-SearchHighlighting)
- [vim-tranquille](https://github.com/RRethy/vim-tranquille)
This PR tries to emulate this behavior natively keeping up with Zed's
sane defaults and deviating from vanilla Vim when it makes sense.
Release Notes:
- Vim: `*` and `#` search doesn't jump immediately to next / previous
search.
This PR improves the visual test infrastructure:
- Adds controllable foreground executor for deterministic task
scheduling
- Adds tooltip hover testing capability
- Improves error handling in visual test runner
- Includes planning documentation for the approach
Release Notes:
- N/A
This PR adds the foundation for the subagents feature:
- Add `SubagentsFeatureFlag` with staff-disabled default
- Create `SubagentTool` struct with input schema for
task/summary/context-low prompts
- Register `SubagentTool` conditionally when feature flag is enabled
- Tool returns stub message 'not yet implemented' for now
Release Notes:
- N/A
---------
Co-authored-by: Amp <amp@ampcode.com>
Co-authored-by: Mikayla Maki <mikayla@zed.dev>
Co-authored-by: Yara <yara@zed.dev>
This builds on https://github.com/zed-industries/zed/pull/46112 (which
should be merged first) and adds the permission evaluation logic and
integrates it with the terminal tool as part of the tool permissions
feature. This is a separate PR for [stacked
diffs](https://newsletter.pragmaticengineer.com/p/stacked-diffs) to make
review easier.
## Changes
- **Add `tool_permissions.rs`** with `decide_permission()` function that
implements:
- `deny > confirm > allow` precedence hierarchy (security-critical)
- Case-sensitive and case-insensitive regex matching
- Integration with existing `always_allow_tool_actions` setting
- Comprehensive unit tests (15 tests)
- **Integrate with terminal tool**:
- Commands matching deny rules are blocked immediately with an error
- Commands matching allow rules proceed without confirmation dialog
- Commands requiring confirmation show the dialog as before
- Added integration tests for deny and allow rule scenarios
Co-Authored-By: Claude Opus 4.5
Release Notes:
- N/A
---------
Co-authored-by: Amp <amp@ampcode.com>
When focus is on the conversation part of the agent panel (not the
message editor), pressing Escape now interrupts the running thread.
Previously, Escape only worked when the message editor had focus.
Release Notes:
- Pressing Esc when the agent panel is focused now interrupts the
running thread even if the text input box is not specifically focused.
Implements visual tests for breakpoint hover states in the editor gutter
with three states:
1. **No hover** - just line numbers
2. **Hover with indicator** - blue circle appears next to the line
3. **Hover with tooltip** - shows 'Set breakpoint' with F9 keybinding
- Added `VisualTestPlatform` that combines real Mac rendering with
controllable `TestDispatcher` for deterministic task scheduling and time
control
- Added `VisualTestAppContext::with_asset_source()` to enable SVG icon
rendering in visual tests (uses real Assets instead of empty source)
- Added F9 keybinding for ToggleBreakpoint so tooltip shows the
keybinding
- Removed debug `eprintln!` statements from editor, element, svg, and
window
The `VisualTestPlatform` enables `advance_clock()` which is essential
for testing time-based behaviors like tooltip delays. Without
deterministic time control, tests for hover states and tooltips would be
flaky.
Release Notes:
- N/A
This PR adds some clean up to the worktree picker: making sure that
worktree names render just one line, even when they might have multiple
lines (that was making each line item break for me), removing
superfluous icons, and making the footer controls more consistent with
the branch picker, where the buttons change to "Create..." if you're in
a state where the only match is the one to create a new worktree.
https://github.com/user-attachments/assets/e63aa8c9-b5a0-443d-a2ab-7759daed10d1
Release Notes:
- Git: Fixed some UI bugs in the worktree picker as well as improved the
UX for creating worktrees based off of the current or default branch.
I really enjoyed this feature in Claude Code. Helps me get a sense of
how effortful something is.
Release Notes:
- Added a "show_turn_stats" setting, default to false, that shows the
timer and the number of tokens down.
This makes running `predict` with the teacher model much faster, when
there are many examples.
Release Notes:
- N/A
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Tackling this as I noticed a bug in the agent panel where the button to
delete a thread, which appeared only on hover, stopped showing up. PRs
#43324 and #45437 fixed stuff in applying hover styles through
`.hover()` but broke the `.on_hover()` callback. Problem was that both
methods were sharing the same `element_state.hover_state` but running at
different phases. The solution here was to add a new independent state
field for the hover listener (`hover_listener_state`) while the hover
style method keeps using `hover_state`.
Release Notes:
- Agent: Fixed a bug where the button to delete a thread stopped showing
up.
This PR removes the const generics limitation from the data table
component, enabling tables with a dynamic number of columns determined
at runtime.
**Context:**
This is the next infrastructure step split out from the [original CSV
preview draft PR](https://github.com/zed-industries/zed/pull/44344). The
draft PR remains open as a reference and will continue to be decomposed
into smaller, reviewable pieces.
**Details:**
- Introduces a `TableRow` newtype to enforce column count invariants at
runtime, replacing the previous const generic approach. It's api surface
is larger than is currently used, as it's planned to be used in CSV
feature itself.
- Refactors the data table and all usages (including the keymap editor)
to work with runtime column counts.
- This change is foundational for supporting CSV preview and other
features that require flexible, dynamic tables.
- Performance impact has not been formally measured, but there is no
noticeable slowdown in practice.
---
Release Notes:
- N/A (internal infrastructure change, no user impact)
---------
Co-authored-by: Anthony Eid <anthony@zed.dev>
Some code got added to `workspace` that prevents us from running tests
for the `edit_prediction(cli)` crates specifically without the
`test-support` feature flag.
Release Notes:
- N/A
Co-authored-by: Ben Kunkle <ben@zed.dev>
Closes#45776
It didn't enable multiline mode before `¯\_(ツ)_/¯`. If there is a reason
this fix won't work please let me know but it seems like this one was
just an easy fix.
Release Notes:
- Improved multiline regex behavior
Before this we failed to parse some file:/// links the agent would
generate causing it to open in the system default app and not zed.
Release Notes:
- N/A
If I apply
```diff
diff --git a/crates/action_log/src/action_log.rs b/crates/action_log/src/action_log.rs
index 404fb3616d..ece063c34f 100644
--- a/crates/action_log/src/action_log.rs
+++ b/crates/action_log/src/action_log.rs
@@ -223,6 +223,7 @@ impl ActionLog {
futures::select_biased! {
buffer_update = buffer_updates.next() => {
if let Some((author, buffer_snapshot)) = buffer_update {
+ dbg!(&author);
Self::track_edits(&this, &buffer, author, buffer_snapshot, cx).await?;
} else {
break;
```
on top of `main`, `User` and `Agent` will always interleave.
This happens because `action_log` does updates on `Entity<Buffer>` which
is a current editor's buffer, tracked, and updated by agent output (acp
or regular threads) — those updates come back as `BufferEvent::Edited`
event after each agent's edit and forces unnecessary computations.
Instead, update tracked buffer's version after each agent update report
to only react on one, `Agent`-authored, edit events.
Release Notes:
- N/A
When a recorded action moves focus away from the editor (e.g.,
`buffer_search::Deploy`), the `EndRepeat` action handler is not invoked
because is node is no longer on the dispatch path. This left
`dot_replaying` set to `true`, causing subsequent repeats to malfunction
and the `VimGlobals.pre_count` value to never be reset.
Reset `dot_replaying` as a fail-safe when the replayer exhausts its
action queue, ensuring the state is always cleaned up regardless of
whether `EndRepeat` was handled.
Release Notes:
- Fixed vim repeat (`.`) breaking when the recorded action moves focus
away from the editor
Co-authored-by: neel <neel@chot.ai>
The `vim` crate's tests depend on `git_ui`, which transitively depends
on `recent_projects` with `test-support` enabled. This causes
`recent_projects` to include RemoteConnectionOptions::Mock` variant
handling. However, `git_ui` was not enabling its `test-support` feature,
causing compilation failures when the Mock variant was expected but not
available.
This commit enables the `test-support` feature for both `git_ui` and
`title_bar` dev-dependencies in the `vim` crate, ensuring the Mock
variant is consistently available during testing.
Release Notes:
- N/A
<img width="707" height="778" alt="Screenshot 2026-01-07 at 8 34 00 PM"
src="https://github.com/user-attachments/assets/59842820-079b-4d47-9bdd-f77300f8a60e"
/>
When the user presses Esc or the Stop button to interrupt a thread,
terminal tools now capture their output and include it in the tool
result. This allows the model to see what was happening in the terminal
when the user interrupted, so it can answer questions about the output.
## Changes
- `Thread::cancel()` now returns a `Task` that waits for tools to
respond to cancellation before flushing pending messages
- Terminal tool uses `select!` to detect cancellation signal and
immediately kills the terminal and captures output
- `run_turn_internal` uses `select!` to break out of event loop on
cancel
- Added test for terminal tool cancellation output capture
This is a follow-up to #46218 which added similar functionality for the
"Stop" button on individual terminal tool cards.
Release Notes:
- Interrupting the agent now captures terminal output so the model can
see what was running when you stopped it
Adds a new `ep split` command that splits JSONL datasets into multiple
output files with stratification by `repository_url` when present.
Example usage:
ep split input.jsonl train.jsonl=80% valid.jsonl=rest
Release Notes:
- N/A
If the LSP version label ends up being super long, we now truncate it
and add a tooltip so you can see the whole thing. This avoids the
submenu width from being extremely big:
<img width="500" height="310" alt="Screenshot 2026-01-08 at 9 10@2x"
src="https://github.com/user-attachments/assets/4729b3f8-000d-4bb5-8908-dc243a2d6e7b"
/>
Release Notes:
- Fixed a visual bug where long LSP version labels would increase the
LSP submenu width too much.
Closes#44038
BM25 results were fully replacing fuzzy results, which caused some items
to disappear from search. For example, "Show Menus" wouldn’t show up
when searching for "menu". This happened due to limitations in the
Porter stemmer used by the library. Fuzzy search matched correctly, but
its results were getting overwritten. We now merge both result sets
instead.
Release Notes:
- Fixed an issue where settings search missed some results, like "Show
Menus" when searching for "menu".
Un-preview the active tab when closing other tabs.
In my mind the intent when clicking `Close Others` is to keep the
current tab around, so it should no longer be treated as a preview.
This also matches the VSCode behavior.
Release Notes:
- Make preview tab permanent after closing all other tabs
Fixes#33958
## Problem
PR #44835 attempted to fix SHLVL starting at 2 by removing it from the
`env` HashMap passed to alacritty. However, that HashMap is only an
**overlay** — alacritty uses the parent process environment as a base
and applies the overlay on top. Since alacritty never calls
`env_remove("SHLVL")`, the terminal shell still inherits `SHLVL` from
Zed's process environment.
## Root Cause
The real issue is in `load_login_shell_environment()`:
1. `shell_env::capture()` spawns a login shell to capture environment
variables
2. That login shell increments `SHLVL` (from 0→1 or n→n+1)
3. The captured env (including the incremented `SHLVL`) is written to
Zed's **process environment** via `env::set_var`
4. When you open a terminal, the shell inherits Zed's `SHLVL` and
increments it again → starts at 2
5. On reload, `shell_env::capture()` runs again with the
already-elevated `SHLVL`, incrementing it further → +2 per reload
## Fix
Skip `SHLVL` when setting Zed's process environment in
`load_login_shell_environment()`. This prevents the login-shell capture
from polluting Zed's env, so terminals start fresh with the correct
`SHLVL`.
Release Notes:
- Improved Display the actual Windows error message when writing
credentials to Credential Manager fails, instead of the generic "Failed
to write API key to keychain" message. This helps users diagnose issues
like permission problems.
Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Also improves the commit view to use the same status-based buffer header
visual overrides as the project diff as a driveby.
Fixes https://github.com/zed-industries/zed/issues/45735
Release Notes:
- git: Binary files are no longer shown in garbled form when viewing an
old commit.
Hi Zed team thank you for the awesome editor!
I recently stumbled upon a markdown sequence that seems to cause a crash
```md
- [ ] -
\
-
```
*easier to read escape characters below
```rust
let crash_input = "-\t[\t] -\r\\\n-"
println!("{}", crash_input)
```
## how to reproduce
1. copy the markdown above
2. save the file
3. `[shift]` + `[cmd]` + `p` to open the command palette
4. select `markdown: open preview`
5. crash
I've confirmed that the issue is a bug in pulldown-cmark version 12, and
has been resolved in version
[v0.13.0](https://github.com/pulldown-cmark/pulldown-cmark/releases/tag/v0.13.0)
and specifically fixed in
https://github.com/pulldown-cmark/pulldown-cmark/pull/1017
this PR simply bumps the pulldown-cmark version in zed which resolves
the crash on my local machine.
## recording
https://github.com/user-attachments/assets/dc77132f-0d43-40f3-9841-0bf34fe714fb
Release Notes:
- Fixes crash due to markdown parsing via bumping pulldown-cmark
Richard Feldman