0.8.3

Added

• Right-click a gutter icon to open the CI report. Left-click still opens the local report in a single click; when a CI report URL template is configured (Settings → Tools → Kensa), the gutter icon's right-click menu now offers "Open Local Report" and "Open CI Report", and its tooltip advertises the CI option. Previously the local/CI chooser lived only on the Alt+Enter intention menu, which devs rarely discovered, so a configured CI template appeared to do nothing — the gutter always jumped straight to the local report. KensaGutterLineMarkerProvider supplies the chooser through the gutter renderer's getPopupMenuActions(), so the fast local path is unchanged.

Fixed

• High idle CPU. KensaOutputFileWatcherStartupActivity polled for new reports by walking the entire project tree (including build/, .git, node_modules) every 5 seconds, forever — continuous filesystem I/O even when the IDE was idle. The walk existed because reports written by terminal/external test runs land under an excluded build/ dir that VFS never reports. It is replaced by three bounded layers: (1) a native file-watch (addRootToWatch) on each discovered report dir, so changes to known reports surface instantly with no polling; (2) a 3-second targeted probe of only the build/output dirs — read from the module model's excluded roots (Gradle build, Maven target) — checking the fixed report shape (<buildDir>/<reportDir>/indices.json or …/sources/<id>/indices.json) so brand-new reports appear quickly without ever walking a build dir's contents; (3) a 60-second pruned safety walk for reports in unconventional locations, which also refreshes the build-root set and prunes stale entries. Idle CPU drops to near zero while a terminal test run still updates gutter icons within ~3 seconds (instantly for already-known reports). Site-mode bundles (build/<site>/sources/<id>/) are handled across all three layers.

0.8.2

Fixed

• Run-window "Open Kensa Report" toolbar icon now appears reliably after Gradle-delegated test runs. KensaTestRunListener captured the run tab's RunContentDescriptor in onTestingStarted — too early for Gradle delegation, where the Build window holds focus while Gradle ramps up, so classes were filed under a stale or empty descriptor that the toolbar's update() never resolved. The listener now accumulates the run's classes on the stable test-root proxy and binds them to the descriptor once, at onTestingFinished, when the test tab is reliably the selected content.
• Toolbar icon no longer renders as a giant full-width button that opens the report when clicked anywhere on the toolbar. The shared action icon (/icons/logo.svg) declared a 512×512 intrinsic size, so IntelliJ sized the ActionButton to fill the whole toolbar; it is now a normal 16×16 action icon.
• Opening a report from the run-window toolbar icon or the test-tree menu now selects the test in single-source (non-site) reports. buildKensaRoute omitted the source prefix for single-source bundles, producing #/test/<class>, but the report keys class nodes as <sourceId>::<class> and falls back to source id default; the route now always carries the prefix, so the report lands on the test instead of opening unselected.

0.8.2-eap.4

Fixed

• Run-window "Open Kensa Report" toolbar icon now appears reliably after Gradle-delegated test runs. KensaTestRunListener captured the run tab's RunContentDescriptor in onTestingStarted — too early for Gradle delegation, where the Build window holds focus while Gradle ramps up, so classes were filed under a stale or empty descriptor that the toolbar's update() never resolved. The listener now accumulates the run's classes on the stable test-root proxy and binds them to the descriptor once, at onTestingFinished, when the test tab is reliably the selected content.
• Toolbar icon no longer renders as a giant full-width button that opens the report when clicked anywhere on the toolbar. The shared action icon (/icons/logo.svg) declared a 512×512 intrinsic size, so IntelliJ sized the ActionButton to fill the whole toolbar; it is now a normal 16×16 action icon.

0.8.2.eap.1

Fix version number in CHANGELOG.md

0.8.2-eap.3

Fixed

• Run-window "Open Kensa Report" toolbar icon now appears reliably after Gradle-delegated test runs. KensaTestRunListener captured the run tab's RunContentDescriptor in onTestingStarted — too early for Gradle delegation, where the Build window holds focus while Gradle ramps up, so classes were filed under a stale or empty descriptor that the toolbar's update() never resolved. The listener now accumulates the run's classes on the stable test-root proxy and binds them to the descriptor once, at onTestingFinished, when the test tab is reliably the selected content.
• Toolbar icon no longer renders as a giant full-width button that opens the report when clicked anywhere on the toolbar. The shared action icon (/icons/logo.svg) declared a 512×512 intrinsic size, so IntelliJ sized the ActionButton to fill the whole toolbar; it is now a normal 16×16 action icon.

0.8.2-eap.2

Fixed

• Investigating why test window toolbar icon is not always displayed for Kensa tests.

0.8.1

Added

• Site-mode discovery: gutter icons and the test-tree context menu now recognise Kensa 0.8 site bundles (build/kensa-site/sources/<id>/) in addition to the default kensa-output/ layout. The plugin walks for indices.json whose grandparent is sources/ and whose great-grandparent has a manifest.json.
• Multi-source routing: when a test class lives in a Gradle source set whose name matches a Kensa source id (the default mapping), gutter clicks open the site shell with a <sourceId>::<class> route, so the report sidebar lands on the correct source. Same class run in two source sets stays separated; the gutter icon for a file in src/uiTest/... reflects the uiTest source's status.
• Tools → "Install Kensa Agent Skills…" action (also surfaced in Settings → Tools → Kensa) that writes the kensa-development AI skill into a project. Targets: GitHub Copilot (path-scoped or always-loaded), JetBrains Junie, Cursor, and Claude Code. Skill files are bundled with the plugin and pinned to the Kensa version in version.txt (sourced from kensa-dev/agent-skills).

Fixed

• KensaTestRunListener failed to instantiate on IntelliJ 2026.1+ (build 261+) with Cannot find suitable constructor, breaking gutter status updates and the engagement notification on that build line. Added @JvmOverloads so the platform's stricter constructor lookup finds the (Project) overload.
• Possible fixes for unreliable display of test runner icon & Kensa Report bubble.
• Opening a report from a site-mode source bundle whose shell isn't on disk now produces a clear "run ./gradlew assembleKensaSite" warning, instead of silently failing.
• Project-view "Open Kensa Report" group now lists site-mode reports (build/kensa-site/index.html), not just kensa-output/ bundles. Previously the group's children walker only matched the legacy parent name and rendered the menu disabled in site-mode projects.
• "Kensa Report Ready" balloon no longer fires twice when a multi-sourceset site build completes. The notification debounce is now per-index.html over a 60s window, so the two sources/<id>/indices.json writes that share one site shell coalesce into a single balloon.
• Run-tab "Open Kensa Report" toolbar icon stayed hidden after Gradle-delegated test runs. KensaTestRunListener.onTestingStarted was capturing the active RunContentDescriptor via RunContentManager.selectedContent, but for Gradle-delegated runs the Build tool window holds focus while Gradle ramps up — selected content is null (or the previous tab) at that moment, so DESCRIPTOR_KEY never gets written, rootDescriptor() returns null on every later event, and KensaRunTabRegistry.seenClasses stays empty for the whole session. The listener now lazy-captures: if DESCRIPTOR_KEY is unset when an event arrives, it reads selectedContent then and stores it back.
• Clicking the run-tab toolbar icon now opens the report routed to a test from the just-completed run (using the recorded class's sourceId), matching the gutter / test-tree right-click behaviour, instead of landing on the UI's default sidebar selection.
• Site-mode discriminator no longer requires manifest.json to be on disk — it matches the directory shape …/kensa-site/sources/<id>/indices.json, so reports load after running :uiTest standalone without :assembleKensaSite. The "open" path still surfaces the missing-shell warning when the user clicks before assembling the site.

0.8.1-eap.1

Added

• Site-mode discovery: gutter icons and the test-tree context menu now recognise Kensa 0.8 site bundles (build/kensa-site/sources/<id>/) in addition to the default kensa-output/ layout. The plugin walks for indices.json whose grandparent is sources/ and whose great-grandparent has a manifest.json.
• Multi-source routing: when a test class lives in a Gradle source set whose name matches a Kensa source id (the default mapping), gutter clicks open the site shell with a <sourceId>::<class> route, so the report sidebar lands on the correct source. Same class run in two source sets stays separated; the gutter icon for a file in src/uiTest/... reflects the uiTest source's status.
• Tools → "Install Kensa Agent Skills…" action (also surfaced in Settings → Tools → Kensa) that writes the kensa-development AI skill into a project. Targets: GitHub Copilot (path-scoped or always-loaded), JetBrains Junie, Cursor, and Claude Code. Skill files are bundled with the plugin and pinned to the Kensa version in version.txt (sourced from kensa-dev/agent-skills).

Fixed

• KensaTestRunListener failed to instantiate on IntelliJ 2026.1+ (build 261+) with Cannot find suitable constructor, breaking gutter status updates and the engagement notification on that build line. Added @JvmOverloads so the platform's stricter constructor lookup finds the (Project) overload.
• Possible fixes for unreliable display of test runner icon & Kensa Report bubble.
• Opening a report from a site-mode source bundle whose shell isn't on disk now produces a clear "run ./gradlew assembleKensaSite" warning, instead of silently failing.
• Project-view "Open Kensa Report" group now lists site-mode reports (build/kensa-site/index.html), not just kensa-output/ bundles. Previously the group's children walker only matched the legacy parent name and rendered the menu disabled in site-mode projects.
• "Kensa Report Ready" balloon no longer fires twice when a multi-sourceset site build completes. The notification debounce is now per-index.html over a 60s window, so the two sources/<id>/indices.json writes that share one site shell coalesce into a single balloon.
• Run-tab "Open Kensa Report" toolbar icon stayed hidden after Gradle-delegated test runs. KensaTestRunListener.onTestingStarted was capturing the active RunContentDescriptor via RunContentManager.selectedContent, but for Gradle-delegated runs the Build tool window holds focus while Gradle ramps up — selected content is null (or the previous tab) at that moment, so DESCRIPTOR_KEY never gets written, rootDescriptor() returns null on every later event, and KensaRunTabRegistry.seenClasses stays empty for the whole session. The listener now lazy-captures: if DESCRIPTOR_KEY is unset when an event arrives, it reads selectedContent then and stores it back.
• Clicking the run-tab toolbar icon now opens the report routed to a test from the just-completed run (using the recorded class's sourceId), matching the gutter / test-tree right-click behaviour, instead of landing on the UI's default sidebar selection.
• Site-mode discriminator no longer requires manifest.json to be on disk — it matches the directory shape …/kensa-site/sources/<id>/indices.json, so reports load after running :uiTest standalone without :assembleKensaSite. The "open" path still surfaces the missing-shell warning when the user clicks before assembling the site.

0.8.0

Added

• Site-mode discovery: gutter icons and the test-tree context menu now recognise Kensa 0.8 site bundles (build/kensa-site/sources/<id>/) in addition to the default kensa-output/ layout. The plugin walks for indices.json whose grandparent is sources/ and whose great-grandparent has a manifest.json.
• Multi-source routing: when a test class lives in a Gradle source set whose name matches a Kensa source id (the default mapping), gutter clicks open the site shell with a <sourceId>::<class> route, so the report sidebar lands on the correct source. Same class run in two source sets stays separated; the gutter icon for a file in src/uiTest/... reflects the uiTest source's status.
• Tools → "Install Kensa Agent Skills…" action (also surfaced in Settings → Tools → Kensa) that writes the kensa-development AI skill into a project. Targets: GitHub Copilot (path-scoped or always-loaded), JetBrains Junie, Cursor, and Claude Code. Skill files are bundled with the plugin and pinned to the Kensa version in version.txt (sourced from kensa-dev/agent-skills).

Fixed

• KensaTestRunListener failed to instantiate on IntelliJ 2026.1+ (build 261+) with Cannot find suitable constructor, breaking gutter status updates and the engagement notification on that build line. Added @JvmOverloads so the platform's stricter constructor lookup finds the (Project) overload.
• Possible fixes for unreliable display of test runner icon & Kensa Report bubble.
• Opening a report from a site-mode source bundle whose shell isn't on disk now produces a clear "run ./gradlew assembleKensaSite" warning, instead of silently failing.
• Project-view "Open Kensa Report" group now lists site-mode reports (build/kensa-site/index.html), not just kensa-output/ bundles. Previously the group's children walker only matched the legacy parent name and rendered the menu disabled in site-mode projects.
• "Kensa Report Ready" balloon no longer fires twice when a multi-sourceset site build completes. The notification debounce is now per-index.html over a 60s window, so the two sources/<id>/indices.json writes that share one site shell coalesce into a single balloon.
• Run-tab "Open Kensa Report" toolbar icon stayed hidden after Gradle-delegated test runs. KensaTestRunListener.onTestingStarted was capturing the active RunContentDescriptor via RunContentManager.selectedContent, but for Gradle-delegated runs the Build tool window holds focus while Gradle ramps up — selected content is null (or the previous tab) at that moment, so DESCRIPTOR_KEY never gets written, rootDescriptor() returns null on every later event, and KensaRunTabRegistry.seenClasses stays empty for the whole session. The listener now lazy-captures: if DESCRIPTOR_KEY is unset when an event arrives, it reads selectedContent then and stores it back.
• Clicking the run-tab toolbar icon now opens the report routed to a test from the just-completed run (using the recorded class's sourceId), matching the gutter / test-tree right-click behaviour, instead of landing on the UI's default sidebar selection.
• Site-mode discriminator no longer requires manifest.json to be on disk — it matches the directory shape …/kensa-site/sources/<id>/indices.json, so reports load after running :uiTest standalone without :assembleKensaSite. The "open" path still surfaces the missing-shell warning when the user clicks before assembling the site.

0.8.0-eap.1

Added

• Site-mode discovery: gutter icons and the test-tree context menu now recognise Kensa 0.8 site bundles (build/kensa-site/sources/<id>/) in addition to the default kensa-output/ layout. The plugin walks for indices.json whose grandparent is sources/ and whose great-grandparent has a manifest.json.
• Multi-source routing: when a test class lives in a Gradle source set whose name matches a Kensa source id (the default mapping), gutter clicks open the site shell with a <sourceId>::<class> route, so the report sidebar lands on the correct source. Same class run in two source sets stays separated; the gutter icon for a file in src/uiTest/... reflects the uiTest source's status.
• Tools → "Install Kensa Agent Skills…" action (also surfaced in Settings → Tools → Kensa) that writes the kensa-development AI skill into a project. Targets: GitHub Copilot (path-scoped or always-loaded), JetBrains Junie, Cursor, and Claude Code. Skill files are bundled with the plugin and pinned to the Kensa version in version.txt (sourced from kensa-dev/agent-skills).

Fixed

• KensaTestRunListener failed to instantiate on IntelliJ 2026.1+ (build 261+) with Cannot find suitable constructor, breaking gutter status updates and the engagement notification on that build line. Added @JvmOverloads so the platform's stricter constructor lookup finds the (Project) overload.
• Possible fixes for unreliable display of test runner icon & Kensa Report bubble.
• Opening a report from a site-mode source bundle whose shell isn't on disk now produces a clear "run ./gradlew assembleKensaSite" warning, instead of silently failing.
• Project-view "Open Kensa Report" group now lists site-mode reports (build/kensa-site/index.html), not just kensa-output/ bundles. Previously the group's children walker only matched the legacy parent name and rendered the menu disabled in site-mode projects.
• "Kensa Report Ready" balloon no longer fires twice when a multi-sourceset site build completes. The notification debounce is now per-index.html over a 60s window, so the two sources/<id>/indices.json writes that share one site shell coalesce into a single balloon.
• Run-tab "Open Kensa Report" toolbar icon stayed hidden after Gradle-delegated test runs. KensaTestRunListener.onTestingStarted was capturing the active RunContentDescriptor via RunContentManager.selectedContent, but for Gradle-delegated runs the Build tool window holds focus while Gradle ramps up — selected content is null (or the previous tab) at that moment, so DESCRIPTOR_KEY never gets written, rootDescriptor() returns null on every later event, and KensaRunTabRegistry.seenClasses stays empty for the whole session. The listener now lazy-captures: if DESCRIPTOR_KEY is unset when an event arrives, it reads selectedContent then and stores it back.
• Clicking the run-tab toolbar icon now opens the report routed to a test from the just-completed run (using the recorded class's sourceId), matching the gutter / test-tree right-click behaviour, instead of landing on the UI's default sidebar selection.
• Site-mode discriminator no longer requires manifest.json to be on disk — it matches the directory shape …/kensa-site/sources/<id>/indices.json, so reports load after running :uiTest standalone without :assembleKensaSite. The "open" path still surfaces the missing-shell warning when the user clicks before assembling the site.