Closes#969
## What and why
On Windows, deleting an agent task and its worktree could fail with `os error 32` while the agent's process tree still held the directory or its files open. Deletion now waits up to 5 seconds for the agent process to be killed and reaped, then retries folder removal for transient locks. This also covers persistent ACP agents such as CodeBuddy that cannot reload their session and therefore stay alive while idle.
## How it was verified
- Windows desktop acceptance through the Tauri MCP bridge, using a deterministic ACP agent with a real PowerShell descendant holding a file open. Closed the conversation tab, then deleted the task with worktree removal enabled. A separate helper retained another file lock for 1.2 seconds: the pre-fix implementation showed `os error 32`; the fixed implementation deleted the task and directory, with both agent processes gone.
- Also verified direct deletion, archive then delete, and deletion with worktree removal unchecked. Processes exited in all cases; branches were preserved, and unchecked removal preserved the folder.
- `cargo test -p ide-jsonrpc -p ide-agent -- --test-threads=2`: 155 passed, 5 real-agent tests ignored. The test runner used Git Bash on PATH and a process-local `core.autocrlf=false`, matching CI's LF fixture setup; no global Git settings were changed.
- `cargo clippy -p ide-jsonrpc -p ide-agent --all-targets -- -D warnings`, `cargo fmt --all --check`, and `git diff --check`: passed.
- Full workspace (Windows/macOS/Linux), frontend and E2E CI: [final PR run](https://github.com/l0ng-ai/openidea/actions/runs/41819945536).
- Windows regression coverage now checks deletion while waiting for permission and while idle without `session/load`, actual sharing violations, delayed file-lock release, and `Peer::exited()` waiting for a descendant's file handle to close. Existing Unix process tests remain enabled.
- Evidence (local, not committed):
- Before: `C:/Users/thoma/repo/pr1240-evidence-20261104/before.png`
- After: `C:/Users/thoma/repo/pr1240-evidence-20261104/after.png`
- Recording: `C:/Users/thoma/repo/pr1240-evidence-20261104/after.gif`
- Runtime results: `C:/Users/thoma/repo/pr1240-evidence-20261104/runtime-cases.json`
- Real signed-in CodeBuddy and native Claude/Codex were not exercised. Native Claude/Codex use an in-process connection, so their `exited()` resolves immediately and removal relies on the bounded retry. Shared-worktree fork protection remains unchanged.
Settings › Tools › Agents › Built-in has an Environment variables field under the executable. They are saved in the agent's file like a command override, and the agent starts with them on top of the shell environment, replacing variables of the same name. Changing the command or the environment keeps the other; a file left with only the built-in's values is removed.
Closes#1133
A task's agent process lived until the task was archived or the project closed, and stopping it killed only the direct child, so wrappers (npx, codex.cmd, claude shims) left the real agent running. Agents that support session/load are now stopped after 5 minutes idle and restarted with the session reloaded on the next message; agent connections, language servers and the native Claude CLI start in their own process group and are stopped with their whole tree.
New tasks run in the local checkout unless the user picks a worktree, and worktrees go next to the project (<parent>/<project>.worktrees/<task>), on the project's drive and outside its tree.
Fixes#1011Fixes#1003
Claude Code reports the account's 5-hour and weekly usage limits (rate_limit_event) and the session cost (result.total_cost_usd); only the limit status was used. The limits now show as meters next to the context meter in the task header, with the reset time in the tooltip, and Settings › Tools › Agents › Chat turns each meter (context, work time, limits, cost) on or off. Limits are per provider and kept in memory; cost is off by default.
Closes#1057
Co-authored-by: matthiasdemauroy <295796358+matthiasdemauroy@users.noreply.github.com>
Settings › Tools › Agents is split into Built-in and ACP Registry tabs. The Built-in tab lists every built-in agent with its status (custom executable or not found on PATH) and, for the selected agent, its default launch line, an Executable field (empty = auto-detect on PATH) with Browse and Reset, and its sign-in hint. Edits are staged with the Settings dialog and written on Apply / OK; Cancel discards them.
The override is written to <config>/agents/<id>.toml, keeping hand-written name, driver, args and env; resetting a command-only override removes the file. AgentProvider now exposes the program, args, built-in defaults, login hint and whether the command is editable: built-ins started through a package runner (npx) and registry installs are read-only, with the reason shown.
Fixes#596
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Deleting a worktree-isolated agent task together with its worktree failed for good with "fatal: '<path>' is not a working tree" once git no longer recognized the worktree: its registration was pruned, the repository moved, or the worktree's `.git` file was deleted.
When `git worktree remove --force` fails but the folder is still there and lies inside the agent worktrees directory, the folder is now removed directly, so the task can always be deleted. A stale registration is pruned by git later. Paths outside that directory (the task's cwd comes from its stored record) keep git's error and are never deleted by hand.
Fixes#969.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
GitHub Copilot, Qwen Code and Mistral Vibe were listed in the ACP Registry under their registry ids (github-copilot-cli, qwen-code, mistral-vibe), so they showed as unrelated third-party agents and could not be installed over the built-in providers. Registry ids now map to the built-in ids: the rows use the built-in name and icon, install and update into <config>/agents/<id>.toml, show "x.y.z available" when the registry has a newer version, and fall back to the built-in command on PATH after Remove. Installing never overwrites a hand-written agent config, and Remove only deletes a registry install. Changing installed providers reloads them and drops stale probes, so New Task sees the installed version right away.
Co-authored-by: dengliu <5108773+dengliu@users.noreply.github.com>
YAML/properties keys stayed inside the generic YAML language server, so Spring configuration bindings and their consumers were unreachable from the editor. A new command in the bundled JDT LS extension (java-usages 1.2.0) resolves keys to @ConfigurationProperties members (prefixes, nested/inherited members, records, constructor aliases, collections), @Value placeholders and PropertyResolver calls, and back from Java to configuration entries; Find Usages merges configuration consumers and declarations with the native Java results. Several profiles or modules are offered as candidates. Closes#796.
Co-authored-by: qjlqyf <109515705+qjlqyf@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Prompts are sent with the origin `human` and their results echo it, but
`belongs` treated any result with an origin as another turn's. The result
was ignored and the turn ended on `command_lifecycle`, which dropped the
failure the turn had recorded. A signed-out Claude Code showed an empty
reply and a Ready task, with no error.
Only origins other than `human` now mark a result as foreign, so failures,
usage limits and the `max_tokens` and `refusal` stop reasons reach the task.
A turn ended by `command_lifecycle` also keeps its recorded failure.
* build: verify native Windows and Linux ARM64 bundles
* fix: select native ARM64 tools before Windows x64 fallback
* build: register verified ARM64 updater bundles and remove verification workflow
* fix(agent): prefer npx and uvx over emulated x64 binaries on Windows ARM64
* fix(tools): include the Windows user PATH agents and tools are resolved on
On Windows the login-shell environment is never read (it is a Unix-only path),
so a process started before the user PATH was edited — or one that simply does
not inherit it — resolves agent executables and language-server tools on the
machine PATH alone. A tool installed for the user (npm -g, \dotnet tool
install -g\, a vendor installer) is then reported as not on PATH.
Read the user's own PATH from HKCU\\Environment (expanding REG_EXPAND_SZ
entries) and append its directories after the inherited ones in both the tool
search path and the native agent lookup, so user installs resolve without
shadowing the inherited order.
Fixes#330.
* fix(tools): use div_ceil for the registry buffer size
* fix(tools): add the Windows user PATH to the process PATH at startup
Lookups alone left agents reported as not installed, and programs started from
user PATH entries could not find their own dependencies.
---------
Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
Workspaces open several local project folders in one IDE session (File › New / Open / Manage Workspace…, or Attach when opening a project, which creates a workspace automatically for a plain project window). Member folders are the Project view roots and take part in indexing, Find in Files, Git, language servers, run configurations and builds; agent tasks choose which member they work in. Window › Merge All Project Windows groups project windows into one window with project tabs (switch, reorder, close, move to a new window), restored on reopen.
Closes#241Closes#198Closes#286
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.
New agent tasks start in the agent's mode that approves everything without asking (bypassPermissions, yolo, full-access…) or with approvals bypassed, unless the New Task page sets something else.
Agent tasks failed in projects without git because change tracking assumed a repository. Outside a repository there are no changes to compare; worktree isolation is hidden and the task runs in the local checkout, and asking for a worktree there returns a clear error.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Three ide-agent tests occasionally hung forever (about 1 in 40 runs under parallel load). Their fake agent answers a turn by requesting permission from a blocking thread and waits for the reply; the tests ended while that request was pending, and dropping the Tokio runtime waits for its blocking pool, which never finished. The tests now cancel the pending turn and wait for the task to settle before shutting down. Test-only: in the app, shutdown kills agent processes rather than waiting on them.
Archived agent tasks could be opened but never returned to the active list, so their message composer stayed disabled. Unarchive is now available in the task header, in the archived-task notice above the composer, and in the task context menu (including mixed bulk selections). Restoring a task moves it back to the active list, keeps its conversation and worktree, and reconnects the existing agent session on the next message; the backend persists and broadcasts the restored status and refuses restoration while a stopped turn is still settling.
After choosing a model and starting a new chat, the picker still showed that model but the session fell back to the agent's default. The remembered configuration was recorded from the agent's reported defaults before the user's choices were applied, and changing an option during a session did not update it. A new session now remembers its options after applying the user's choices, and changing an option updates the remembered configuration.
Co-authored-by: dimassfeb-09 <62858781+dimassfeb-09@users.noreply.github.com>
The Angular language server (ngserver) was not found when Node and global npm packages came from a version manager (nvm, fnm, volta, asdf, mise) whose bin directories are not on the GUI app's PATH. Tool lookup now searches those directories too through a shared ide-core tool_path module used by the LSP launcher, SDK queries and tool detection; explicit relative commands resolve against the project root, empty PATH entries are ignored, and a non-executable file earlier on PATH no longer shadows the real command.
The process query test that flaked under load now uses a generous deadline for the success case, with the timeout covered by a separate test of a child that never exits.
Co-authored-by: nntndfrk <2458463+nntndfrk@users.noreply.github.com>
OpenCode handles /compact through its ACP session but does not list it in availableCommands, so the composer never offered it. Agent builtins the agent implements without advertising are now merged into slash-command completion without duplicating listed commands; the table holds only commands verified against the agent (OpenCode /compact). Claude Code already advertises /compact. Executable lookup keeps using the shared PATHEXT-aware resolution on main, with a test that the newest install across PATH entries is chosen.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Git blobs were always decoded as UTF-8, so files in GBK or other legacy encodings showed every line as changed in the gutter, diff, blame and commit views, and staging/committing could re-encode them wrongly.
Blob text is now decoded and encoded with the file's encoding: the open document's encoding, else detection on the working file with the project default as fallback, else detection on the blob. Covers line hunks, file-at-revision, blame, stage patch/hunks, commit, conflict versions, diff text, and the agent worktree's file_at, partial commit and patch helpers. Text that cannot be represented in the target encoding is reported as an error instead of being written lossily.
Fixes#243
未在 Windows 实机验证。
On Chinese Windows the Maven tool window showed sync errors as mojibake: tool output was decoded as UTF-8, but while JVMs are told to write UTF-8, cmd.exe, batch launchers (mvn.cmd, gradlew.bat) and native tools write their diagnostics in the ANSI code page (GBK). ide_core::output_encoding::decode_tool_output keeps valid UTF-8 and decodes other lines in the system code page (GetACP). It is used wherever OpenIDEA shows output of tools it launches itself: Maven/Gradle sync and tasks, tool installers, SDK probes, Docker, coverage, environment setup, Cargo stderr for debugging, agent version probes and notebook kernel stderr. Run consoles and protocol streams keep their own encodings. 未在 Windows 实机验证。
Refs #262
On Windows std/tokio Command only tries .exe for a bare program name, so agents installed as npm shims (codex.cmd, claude.cmd) failed with "program not found". Bare names are now resolved through PATH x PATHEXT before spawning.
- ide-core::process gains find_executable / find_executable_in (moved from ide-run::which, which re-exports them).
- ide-jsonrpc Peer::spawn_framed resolves against the child's PATH (caller env, login shell, then ours): covers ACP agents and the native Codex driver.
- ide-agent newest_install builds candidates with executable_names, so the native Claude driver finds .cmd shims and multi-install selection works on Windows.
- ide-agent replay recorder resolves the same way.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
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.
Background tasks often couldn't be cancelled, or cancelling didn't stop them:
- A cancel that arrived before the backend registered the request was dropped (build while saving files, multi-repository fetch/pull between repositories). The host now remembers early cancels (bounded, 2 min), and request ids start at a random value per page.
- There was no queue and no common cancel path: backend-reported tasks (LSP progress, deployment transfers) couldn't be cancelled, and tool installs, VCS operations and update checks had no cancel button in Background Tasks.
- Only the direct child process was killed, and JDK / tool archives were extracted without checking for cancellation.
Tool installs and JDK downloads now share a downloads queue; cancelling a queued task removes it without starting any work. Every task cancels through one path that shows "Cancelling…" and then removes it; tasks that can't be interrupted show a disabled cancel button with the reason. IdeEvent::Task carries `cancellable` with a cancel_task command, cancellable LSP progress sends window/workDoneProgress/cancel, installers, curl and tar run in their own process group and are killed as a group, and extraction checks for cancellation between entries.
Claude-Session: https://claude.ai/code/session_01Dq6jQQn8ZhRfkkyj3PRutA
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
* 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
* fix(agents): Leave changes made before a local checkout task out of its changes
A task in the project's own checkout listed every uncommitted and untracked
file as its own. The files already changed when the task starts are now
recorded with their size and modification time, and drop out of the task's
changes until they are edited again.
* feat(agents): Show the current branch for a local checkout task
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.
The task title starts as the first line of the prompt. When the agent sends
a session_info_update with a title, the task takes that title, saves it and
shows it in the Agents list.
* 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.
`ide-lsp` tests failed intermittently (about 1 run in 4, more under load).
The main cause, requests from the other side starting before earlier notifications, is fixed on main by b5bc150; this adds a regression test for it in `ide-jsonrpc`, documents that `on_notification` must not wait for the remote side, and drops the 200 ms polling workaround in `ide-agent` that papered over the same race for `session/request_permission`.
Test-only races fixed: mock replies are held with a gate instead of fixed sleeps and timing asserts; the indexing-retry test ends indexing after the first reply; the crash test counts Ready/Crashed transitions instead of reading the last status; the mock records a notification and applies its effect under one lock; `eventually` waits up to a deadline; `tcp_server` reserves its port with a bound socket and waits for the server's start marker.
Claude-Session: https://claude.ai/code/session_01VUYsdXA2nkF5v6zsC93dpn
A save that ran after delete (from the turn ending or the agent exiting)
wrote the task file back, so the task returned on the next load. Deleting
now marks the task under the lock that saving holds.
The new-task composer only offered the slash commands the agent sent in its last session, stored once globally, so a project that hadn't run a session (or whose last session ran elsewhere) showed none of its skills; matching was substring-only, so `/trig` didn't find `/triage` either.
Offered commands are now remembered per project root. For Claude Code, skills (`.claude/skills/*/SKILL.md`) and commands (`.claude/commands/**/*.md`) are also discovered on disk, from the working directory up to home plus `~/.claude`, and merged with the agent's list (agent entries win). New `agent_commands` IPC serves the home composer (project root) and session composer (the task's cwd). Matching ranks prefix (also after `plugin:`), then substring, then subsequence. Picking a completion inserts `/name ` and the prompt is sent unchanged.
Not covered: a new worktree of a project that gitignores `.claude/` doesn't contain the project's skills.
Claude-Session: https://claude.ai/code/session_01VUYsdXA2nkF5v6zsC93dpn
@zed-industries/codex-acp is archived; its last release (0.16.0) bundles an
older Codex core that doesn't know newer models, so a config.toml model such as
gpt-6.1-sol fell back to generic metadata with a warning. The maintained adapter,
@agentclientprotocol/codex-acp, runs the Codex App Server from a current
@openai/codex and lists the account's models.
## 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
matthiasdemauroyandmatthiasdemauroy
arookieofcanda-rookie-of-C-language
Deng Liuanddengliu
qiaojianlongandClaude Opus 5.5
arookieofcanda-rookie-of-C-language
DIMAS FEBRIYANTO
Mykhailo Smolinskyiandnntndfrk