* fix(agents): improve chat transcript interactions
* fix(agents): recover images when regeneration fails
* test(agents): use the native copy shortcut on every host
CI / Rust (windows-latest) (push) Canceled after 0s
Language smoke / Language smoke (java) (push) Failing after 7s
Language smoke / Language smoke (typescript) (push) Failing after 4s
Language smoke / Language smoke (kotlin) (push) Failing after 6s
Language smoke / Language smoke (go) (push) Failing after 7s
Language smoke / Language smoke (python) (push) Failing after 3s
CI / Plugins (SDK, official, fixtures) (push) Failing after 50s
Language smoke / Language smoke (rust) (push) Failing after 6s
CI / Remote server (linux musl) (push) Failing after 1m30s
CI / Rust (ubuntu-22.04) (push) Failing after 1m34s
CI / Cargo deny (push) Failing after 54s
The native Claude Code provider turned each AskUserQuestion question into a
permission request with only the option labels, shown in the permission
banner. Descriptions, headers, previews and multi-select were lost, and the
only way to give another answer was Skip, which tells Claude the user didn't
answer.
All questions now go out in one request carrying the full tool input, and
the conversation shows them as a card above the composer: each option with
its description, radio buttons or checkboxes for multi-select, an Other
choice with a text field, the option's preview beside the options on a wide
card (below them on a narrow one, sized to the tallest preview so hovering
never moves the options), number keys to pick, Esc to skip, and one Submit
for every answer.
Ordinary permission requests keep the banner, and Skip answers as before.
The answers go back in Claude Code's own format: question text to answer,
multi-select picks and Other text comma-joined, with an item that contains
the separator or a quote JSON-quoted so it reads back as one item.
Closes#1628
Co-authored-by: NealimeKenna <[email protected]>
* fix(agent): don't release an idle native agent while background tasks are running
A turn's background tasks (Claude Code's run_in_background commands) outlive
the turn: release_when_idle only checked for a running turn and open
permission requests, so five minutes after the turn ended it stopped the
agent process and orphaned whatever still ran, its result never reaching
the conversation.
Claude Code reports its live background tasks in stream-json mode: a
system/background_tasks_changed event carries the full list on every
change. The driver now translates it into an openidea_background_tasks
session update (like openidea_retry), the task keeps the list, and the
idle release waits while it isn't empty: it checks again after the usual
delay, and when the last task is done its report voids the pending
schedule and starts the delay over from then. A (re)started process
starts with no tasks, so a stale list can never hold the agent up.
Signed-off-by: Inf1mum <[email protected]>
* feat(agents): show the agent's background tasks above the composer
A grey bar above the composer lists what the agent runs in the
background (shells, monitors), with how long each has been running and
a Stop button per task. The task now carries the background tasks the
agent reports, with when each was first seen, and clears them when the
process ends. Stop asks Claude Code to end that one task through a new
_openidea/stop_task request; the composer's Stop keeps ending only the
turn (perTaskStopAffordance).
Closes#1788
---------
Signed-off-by: Inf1mum <[email protected]>
Co-authored-by: Inf1mum <[email protected]>
Co-authored-by: NealimeKenna <[email protected]>
Agent task tabs now stay mounted while another tab is active (the five most
recently used per pane). A hidden one skips rendering with
content-visibility: hidden, which keeps its layout and scroll offset, so
showing it again costs a few milliseconds instead of building and laying out
the whole conversation. Hidden tabs don't mark their task read and close Find.
Thinking blocks and tool output now sit in a wrapper that skips off-screen
layout like other entries; on <details> itself WebKit wouldn't expand them,
so they were laid out in full on every resize and reopen.
A permission request line kept the title, "Waiting for your answer" and the answers on one row, so in a narrow agent conversation a long command title squeezed the answers until they were cut off. Title, state and answers now wrap next to the icon: a long title takes its own line and the state and answers go below it, in line with the title.
The meters were fixed-width and the header didn't clip, so in a narrow
conversation the stop, terminal and changes buttons spilled over the
Changes pane. The title now shrinks first, meters that don't fit drop
out whole, and the buttons keep their size.
The read-only previews beside Problems, Show Usages and TODO results are syntax-coloured from one whole-file parse (capped for very large files), so a line inside a block comment or a multi-line string is coloured as such; the pane renders plain first and recolours when the parse lands. SourcePreview takes the colouring as a colorize prop (the editor's previewColors), keeping src/ui free of editor imports, and the tok-* colours live once in global.css instead of a copy per view, with the theme's comment style.
Co-authored-by: qtvbwfn <[email protected]>
Agent replies only styled paragraphs, lists and code, so tables, headings, quotes, rules, images and task lists fell back to browser defaults: a 26px h1, a quote indented 40px, table columns squeezed to a few letters by overflow-wrap. They now get chat-sized styles from tokens; wide tables scroll sideways.
Closes#1273
Co-authored-by: ntbang0901 <[email protected]>
A pending agent question / permission request could only be answered from the banner at the top of the session; the conversation line said "Waiting for your answer" with nothing to click. The line now offers the same answers, and a request without options gets Cancel in both places so the agent isn't left waiting.
Fixes#1148
CI / Rust (ubuntu-22.04) (push) Failing after 1h0m11s
CI / Remote server (linux musl) (push) Failing after 34m34s
CI / Plugins (SDK, official, fixtures) (push) Successful in 6m55s
CI / Cargo deny (push) Successful in 1m32s
CI / Frontend (push) Successful in 26m39s
CI / E2E (push) Failing after 19m40s
CI / Rust (macos-latest) (push) Canceled after 0s
CI / Rust (windows-latest) (push) Canceled after 0s
The permission banner kept its text on one line, so an agent's question (Claude Code's AskUserQuestion arrives as a permission request) was cut off and its answers crowded into the rest of the line. The question now wraps and scrolls past a third of the editor's height, and the answers wrap after it; permission lines in the conversation show up to three lines with the full text as tooltip.
Agent sessions can be opened in the Agents tool window beside the editor (Open in Tool Window / Open in Editor in the task menus, or by dragging between the editor and the tool window); opening in an editor tab stays the default.
Fixes in review:
- A docked conversation kept the previous task's subagent view after switching tasks: the session's local state is now keyed by task.
- Moving a conversation between the editor and the tool window lost its scroll position: position and follow-output are kept per conversation (task or subagent).
- In a short bottom tool window the message area shrank to a few pixels: wide windows place the task list beside the conversation, narrow ones cap the list height.
- Drag and drop and list selection now go through the agents.dockTask / undockTask / openTask actions.
Co-authored-by: a-rookie-of-C-language <[email protected]>
The Changes pane took a fixed column next to every agent conversation, including tasks that changed nothing. It is now shown while the task has changes (so it appears when the agent first edits a file) and toggled from a Changes button with the file count in the task header; once toggled, the choice is kept per task.
Co-authored-by: a-rookie-of-C-language <[email protected]>
Images sent to an agent were dropped from the conversation: the user message kept only text and file chips. User entries now record their images and show them as thumbnails that open full size; composer thumbnails open the same preview. Editing a sent message keeps its images (removable) and rerunning from it sends them again.
content-visibility: auto on <details> kept collapsed Thinking blocks from expanding in WebKit; details elements are now excluded from the off-screen layout skip.
Co-authored-by: dimassfeb-09 <[email protected]>
When Claude Code loses its API connection or is refused, it retries on its own
for up to several minutes. The session showed only "Working…" meanwhile. It
now shows "API request failed, retrying in Ns (attempt 3/10)" with the error
in the tooltip, cleared by the agent's next output or the end of the turn.
Agents can run through native drivers or any ACP agent, speak ACP v1 or v2, and be installed from the ACP Registry. Tasks gain fork, in-place edit-and-rerun, steering and usage-limit handling where the agent supports them.
- Native Claude Code (`claude` stream-json) and Codex (`codex app-server`) drivers, each served to the task machinery as an in-process ACP shim, using the newest install on PATH.
- ACP v2 with v1 fallback, negotiated on initialize; per-agent capabilities (fork, rollback, steer) so the UI only shows supported actions; usage-limit errors put the task in a `limited` status with the reset time.
- Settings › Tools › Agents lists Registry agents (featured first) to install, update and remove; installed agents update when a project opens. "More agents…" in the provider picker opens it. Registry index and downloads are HTTPS-only, release archives are checked against their sha256, and entries without one show "No checksum provided" and ask for confirmation before installing. Package names from the registry can't be passed as options to npm/uv.
Claude-Session: https://claude.ai/code/session_01Dq6jQQn8ZhRfkkyj3PRutA
Dragging the bottom tool window edge over a terminal refit xterm on every
pointer move and sent each intermediate size to the shell, which redrew its
prompt every time. Terminals now refit at most every 100 ms while their host
keeps resizing, plus once when it stops, and the process only learns the size
once it has held for 150 ms. This covers the Terminal tool window and terminals
embedded in Run consoles.
Resizing the width of an agent conversation reflowed every entry, including the
off-screen ones (14-16 ms per move in a long session). Entries now use
content-visibility: auto, so only the visible ones are laid out.
* feat(agents): Show subagents as their own conversations
Agents that support ACP subagent sessions (Claude Code, Codex) now run each
subagent as a child session. Its tool calls and messages no longer mix into
the task's conversation: the parent shows a card per subagent with its state,
tool call count, latest step, task and result, and the card opens the
subagent's own conversation.
- Route session/update by session id; subagent settings and usage stay out
of the task.
- Subagent permission requests are answered in the task's conversation,
named after the subagent.
- Subagent conversations are saved with the task; a running subagent is
marked stopped after a restart.
- A running subagent can be stopped on its own.
* fix(agents): Save a subagent's state when it changes after the turn
A local-checkout task shows the repository's current branch in its header, and
Changes compares against it; both follow branch switches. The header also shows
the time the agent has spent on the task's turns.
A usage update without a cost (Claude sends one for its rate limits) no longer
clears the session cost.
* feat(agents): Select, copy, quote and search the conversation
- Conversation text can be selected and copied.
- Selected text gets a Quote button that replies to it in the composer.
- Messages have Copy on hover; your own messages also have Edit and
Resend and Send Again.
- Code blocks are highlighted and have a Copy button.
- File paths in replies (`src/lib.rs:12`) open in the editor when the
file exists.
- Find in Conversation (⌘F / Ctrl+F) highlights every match and steps
through them, opening collapsed tool output.
- Jump to Bottom appears after scrolling up.
- Your messages show when they were sent and how long the agent took;
a running turn shows its elapsed time.
* feat(agents): Quote a selection from the context menu
* feat(agents): Drop Jump to Bottom from the conversation
* feat(agents): Drop the hover Copy button from agent messages
* fix(agents): Highlight only the find matches on screen
WebKit paints a highlight range inside a collapsed tool output over the
text next to it. Matches in collapsed output still count, and stepping
to one expands it.
* fix(agents): Repaint find highlights as the search changes
WebKit doesn't repaint where a replaced highlight object was, so matches
of an earlier search stayed painted; each highlight is now one object
changed in place.
- Enter that commits an input-method candidate no longer sends the message.
- The composer turns off autocorrect, autocapitalisation and spellcheck, which
popped up correction bubbles over prompts.
- The provider icon in the task header sits inline with the details instead of
on its own line above them.
## What
An Agents feature: hand a task to a coding agent and follow it from the IDE. Agents speak the [Agent Client Protocol](https://agentclientprotocol.com). Built in: Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor, OpenCode, Qwen Code, Kimi CLI, Goose, Auggie, Kiro, Mistral Vibe. You can add your own in `<config>/agents/<id>.toml` (`name`, `command`, `args`, `[env]`). An agent whose command isn't on `PATH` is listed under "Not installed".
- **New Task page.** An editor page with a prompt box. You pick the agent, model, mode and thinking effort, and whether it runs in a new worktree (from a chosen branch) or in the checkout.
- **Agents tool window.** Tasks are grouped by what they need: Needs Input, Failed, Running, Ready to Review, Archived. It also shows a stripe badge, a status bar item and notifications.
- **Session tab.**
- The conversation: Markdown, thinking, tool calls with diff stats, the plan.
- Permission requests appear as an editor banner.
- The changed files are compared against the task's base commit and open in the diff viewer.
- A context-usage meter.
- **Composer.**
- `@` references files from the project index. A referenced file's text is sent when the agent accepts embedded context, otherwise a `resource_link`.
- `/` lists the agent's commands; ↑/↓ recall earlier messages.
- Images can be pasted, dropped or picked; Shift+Tab cycles the mode.
- While the agent works, Enter queues the message and ⌘↵ interrupts the turn and sends it now. Queued messages can be sent now, edited or removed.
- **Send to Agent.** In the editor and Project view context menus. It adds `@file:10-20` references to the message you typed last.
- Landing the work is left to the existing Git features: the worktree is a normal branch.
## How
- **`crates/ide-agent`**: one ACP agent process per task, running in the task's directory.
- It folds `session/update` into a timeline and answers `session/request_permission`.
- It tracks the session's settings: `configOptions`, falling back to the older `modes` / `models` fields. The user's choices are applied again on every new or reloaded session.
- Tasks are saved as JSON per task. Sessions resume with `session/load` without duplicating history, and old task files are upgraded when read.
- **`ide-jsonrpc`**: adds newline-delimited framing (ACP) next to Content-Length (LSP).
- **`ide-vcs`**: adds `numstat_with_worktree` for the changes list.
- **Editor**:
- The Switcher (⌃⇥) lists every open editor, as IDEA does, not just files.
- `EditorKindContribution.restorable` lets a kind's tabs be saved with the editor session, so agent tabs survive reloads.
- **UI**: `Select` popups now size to fit short option details (they used to cut them off).
## Testing
- `cargo test -p ide-agent` (14 tests). They use an in-memory ACP agent and cover permissions, restart with `session/load`, settings, the queue with send-now and stop, and prompt blocks.
- There is also an ignored end-to-end test against the real Claude Code adapter: `OPENIDEA_AGENT_E2E=1 cargo test -p ide-agent real_claude -- --ignored`.
- `pnpm vitest run`: 2052 tests.
- `cargo clippy -p ide-agent -p openidea --all-targets -- -D warnings` and typecheck are clean.
- Driven in the running app with Claude Code:
- A task in a worktree, with permission answers, changes listed and the diff opened.
- A follow-up after restarting the IDE.
- `@` reference with embedded content, `/` commands, Shift+Tab mode changes.
- A queued message and send-now, an image prompt, Send to Agent from a selection.
- The Switcher and restored tabs after a reload.
Not covered yet: only Claude Code was driven end to end. The other agents go through the same ACP path but weren't run here.
Merged on the maintainer's go-ahead. Merge with main: kept both sides in Cargo.toml (ide-agent + ide-clipboard), the feature registry (agents + vim) and zh.json (no duplicate keys). After the merge: typecheck clean, full vitest 268 files / 2247 tests passed, ide-agent/ide-jsonrpc/ide-vcs 144 passed, bindings unchanged after regeneration.
l0ng-ai
NealimeKennaandNealimeKenna
Vadim Lazuk
Qingandqtvbwfn
ntbang0901
arookieofcanda-rookie-of-C-language
arookieofcanda-rookie-of-C-language
DIMAS FEBRIYANTO