Release v5.8.1

CHANGELOG.md

@@ -4,6 +4,15 @@ All notable changes to this project will be documented in this file.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 Versioning follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [5.8.1] — 2026-08-30
+
+### Fixed
+- **Closing the last chat no longer reports an internal error.** Closing a chat that was the only one
+  open, and that had been used, left the IDE logging `Already disposed` from the browser component the
+  chat had just torn down. The chat's card is now hidden before it leaves the tool window, so the layout
+  never lays out a browser that is already gone. Nothing about closing a chat changes for you: the
+  replacement chat still opens as before.
+
 ## [5.8.0] — 2026-08-28
 
 **A chat in the IDE can now be driven from claude.ai or the Claude app**, and keeps running on your own

CODEOWNERS

@@ -1,7 +1,7 @@
 # CODEOWNERS — Claude Code Native
 #
 # GitHub uses this file to auto-request reviews on pull requests.
-# Maintainer: @serialexperimentslainnnn (the local author "Lain" / lain.agent604@passmail.com).
+# Maintainer: @serialexperimentslainnnn (the local author "Lain" / lain@digitalexperiments.dev).
 
 # Default owner for everything in the repo.
 *                       @serialexperimentslainnnn
@@ -15,6 +15,13 @@
 /src/main/kotlin/dev/lain/claudejb/permission/   @serialexperimentslainnnn
 /src/main/kotlin/dev/lain/claudejb/protocol/     @serialexperimentslainnnn
 
+# Two files nothing may edit in passing. CLAUDE.md carries the standing directive over the
+# guard and its tests, and .gitignore is what keeps credentials and local state out of the
+# history in the first place — a quiet line added to either is a change to the rules, not a
+# chore, so both are owned outright.
+/CLAUDE.md              @serialexperimentslainnnn
+/.gitignore             @serialexperimentslainnnn
+
 # A workflow is privileged code: it runs with a token, and it can publish to the Marketplace. A ruleset is
 # the gate that decides what reaches main at all. Both are listed LAST so they win over every earlier glob.
 #

README.md

@@ -43,6 +43,7 @@ Three different things are often confused. All of them are legitimate; they solv
 | Permissions | An inline card per call, plus a deterministic lock that runs before any auto-approval | Handled by the CLI in the terminal | JetBrains' own approvals |
 | Account | Your `claude` subscription or API key | Your `claude` subscription or API key | JetBrains AI credits, your own Anthropic API key, or a Claude Console account |
 | Agents / background tasks | A tab and a transcript per agent; background tasks keep their output | Visible as terminal output | Not applicable |
+| [Remote Control](#remote-control--pick-a-chat-up-from-your-phone) | A button in the chat and a row in the ⚙ menu; the chat you are looking at is the one that goes to your phone | The CLI's own `/remote-control`, typed in the terminal | Not applicable — it does not run your local CLI |
 | Needs the `claude` CLI | Yes — and installs it for you if you do not have it | Yes | No |
 
 Anthropic's [Claude Code [Beta]](https://plugins.jetbrains.com/plugin/27310-claude-code-beta-) is not
@@ -640,6 +641,13 @@ phone button in the chat's button row and a row in the ⚙ menu, neither of whic
 has accepted; a refusal names its reason on the button and in the chat. Permission cards answered from
 the phone close themselves here.
 
+**5.7.0** — the [Sensitive Guard](#security) stops being invisible: it keeps a log of every alert it
+raises, per project, and a **Guard** view in the chat's view row to read it — what matched, what the rule
+saw, the verdict, and what let the call through if anything did. Every rule gets a mode (Enforcing,
+Permissive, Allow All), as does the guard as a whole; **Settings ▸ Claude Code Security** becomes its own
+page, with the three whitelist reaches and the extra credential globs and blocked domains. A shield in
+the chat's button row suspends the guard for a duration you pick. Most settings became per project.
+
 **5.5.0** — a tab and a transcript per agent, with the whole tree one hover away; a single **Workloads**
 diagram of everything running across every chat; background tasks that keep their output after they
 end and survive a restart; a [Git integration](#git) whose write actions are asked of Claude rather than

RELEASE_NOTES.md

@@ -1,3 +1,11 @@
+## v5.8.1 — 2026-08-30
+
+**Closing your last chat no longer raises an internal error.** If the chat you closed was the
+only one open and you had actually used it, the IDE logged an `Already disposed` error while
+tearing the chat's browser down. The chat is now hidden before it is removed, so nothing tries
+to lay out a browser that has already gone. Closing a chat behaves exactly as before otherwise,
+replacement chat included.
+
 ## v5.8.0 — 2026-08-28
 
 **Pick a chat up from your phone.** Remote Control connects a chat running in your IDE to

build.gradle.kts

@@ -28,7 +28,7 @@ plugins {
 }
 
 group = "dev.lain"
-version = "5.8.0"
+version = "5.8.1"
 
 repositories {
     mavenCentral()

docs/FAQ.md

@@ -89,8 +89,11 @@ Because that is where its settings and credentials live. Since 5.5.0 the whole
 settings document is in the IDE's PasswordSafe rather than in
 `.idea/claude-code.xml` — they were per-project plaintext in a file people
 commit, and they include an env block, which is where an API key or a
-credentialed proxy URL ends up. One consequence worth knowing: **the settings are
-now global, not per project.**
+credentialed proxy URL ends up. The scope worth knowing: **since 5.7.0 the
+settings are one document per IDE installation, per project** — two IDEs on the
+same machine keep separate settings, and so do two projects in the same IDE.
+(Between 5.5.0 and 5.7.0 there was a single global document; it is kept as the
+seed a project without settings of its own starts from.)
 
 If the safe cannot be read (a locked KWallet, say), the plugin treats that as a
 failure and refuses to save over it — a failed read is not an empty

docs/TROUBLESHOOTING.md

@@ -73,8 +73,12 @@ Since **5.5.0** the settings live in the IDE's **PasswordSafe**, not in a file.
 The old `.idea/claude-code.xml` is read once, copied into the safe, and deleted
 only after the safe has accepted the copy. Two things follow:
 
-- **Settings are now global**, not per project. The safe is application-wide,
-  which is also the scope these settings actually had.
+- **Settings are kept per IDE installation, per project** since 5.7.0. The safe
+  is application-wide, but each entry is keyed by the IDE's configuration
+  directory and the project, so two IDEs — or two checkouts — do not share a
+  document. The single global document that 5.5.0–5.7.0 wrote is still read as
+  the seed for a project that has none of its own, which is why it is never
+  deleted.
 - **If the safe cannot be read**, the plugin refuses to save over it rather than
   treating a failed read as an empty configuration. On Linux that usually means a
   locked KWallet/keyring — unlock it and restart the IDE. A read that failed

src/main/kotlin/dev/lain/claudejb/ui/ChatTabsPanel.kt

@@ -116,6 +116,7 @@ internal class ChatTabsPanel : JBPanel<ChatTabsPanel>(BorderLayout()), Disposabl
         } else {
             pushChats()
         }
+        tab.component.isVisible = false
         onClosed(tab)
         content.remove(tab.component)
         tab.disposer?.let { Disposer.dispose(it) }

src/main/kotlin/dev/lain/claudejb/util/PluginIdentity.kt

@@ -2,7 +2,7 @@ package dev.lain.claudejb.util
 
 object PluginIdentity {
 
-    const val PLUGIN_VERSION = "5.8.0"
+    const val PLUGIN_VERSION = "5.8.1"
 
     private const val PROJECT_URL = "https://github.com/serialexperimentslainnnn/claude-code-for-jetbrains"
 

src/test/kotlin/dev/lain/claudejb/headless/ChatTabsCloseHeadlessTest.kt

@@ -108,6 +108,30 @@ class ChatTabsCloseHeadlessTest : BasePlatformTestCase() {
         assertEquals(listOf("A"), closed)
     }
 
+    fun `test the last tab is hidden before it leaves the deck, never shown again`() {
+        val a = open("A")
+        tabs.select(a)
+
+        tabs.close(a)
+
+        assertFalse(
+            "closing the only card must not leave the card layout re-showing the torn down chat",
+            a.component.isVisible,
+        )
+        assertNull("and the card must be out of the deck", a.component.parent)
+    }
+
+    fun `test closing the selected tab hides it and shows the survivor`() {
+        val a = open("A")
+        val b = open("B")
+        tabs.select(a)
+
+        tabs.close(a)
+
+        assertFalse(a.component.isVisible)
+        assertTrue(b.component.isVisible)
+    }
+
     fun `test the closed tab is out of the model when close returns`() {
         val a = open("A")
         open("B")