apps/vscode/CHANGELOG.md

Changelog

0.7.5

Patch Changes

• #3453 411572e Thanks @Grapedge! - Highlight matched characters in @ file suggestions and allow folders to be inserted as mentions.

• #3453 411572e Thanks @Grapedge! - Show the image/video picker entry in the @ menu only before a search query is typed, and remove the Browse folders mode.

• #3453 411572e Thanks @Grapedge! - Fix the @ and / suggestion lists jittering when the mouse rests at the scroll edge.

• Updated dependencies [411572e]:

  • @moonshot-ai/kimi-code-sdk@0.20.0

0.7.4

Patch Changes

• #3371 9e88152 Thanks @tpoisonooo! - Fix prompts remaining queued forever after reopening a session.

• #3366 9619277 Thanks @kimi-agent-bot! - Use the Unicode ellipsis "…" in user-facing TUI and VS Code webview text.

0.7.3

Patch Changes

• #3098 75550c5 Thanks @LCZcn96! - Restore live context-window usage updates after switching to the v2 engine.

• #3276 f34b2ec Thanks @Grapedge! - Fix streamed replies occasionally showing every character twice and tool calls appearing in duplicate.

• Updated dependencies [75550c5, d723cc4]:

  • @moonshot-ai/kimi-code-sdk@0.19.2

0.7.2

Patch Changes

• #3079 35befdc Thanks @gaoyuan1223m! - Fix multi-select questions jumping to the next question after only one answer is selected.

• Updated dependencies [3d77620]:

  • @moonshot-ai/kimi-code-sdk@0.19.1

0.7.1

Patch Changes

• #3026 13857f3 Thanks @bj456736! - Show plugin- and file-declared MCP servers as read-only entries in the MCP servers panel.

• Updated dependencies [d833a1a, 61591bc, d833a1a, 13857f3]:

  • @moonshot-ai/kimi-code-sdk@0.19.0

0.7.0

Minor Changes

• #2916 7475c2e Thanks @Grapedge! - Run the extension on the v2 agent engine by default; the interface, sessions, and workflows are unchanged. To roll back, enable the kimi.useAgentCoreV1 setting and reload the window.

Patch Changes

• Updated dependencies [6be2697, 7475c2e, 7475c2e, 7475c2e]:
  • @moonshot-ai/kimi-code-sdk@0.18.0

0.6.9

Patch Changes

• Updated dependencies [c9bfe8b, c212ae9]:
  • @moonshot-ai/kimi-code-sdk@0.17.0

0.6.8

Patch Changes

• Updated dependencies [437a1b8, 0b2e803, 3c9e3b2]:
  • @moonshot-ai/kimi-code-sdk@0.16.0

0.6.7

Patch Changes

• #2326 302b2cd Thanks @gaoyuan1223m! - Fix only the first question being answerable when the agent asked multiple questions at once; each question is now answered one by one and submitted together.

0.6.6

Patch Changes

• #2393 6d0a046 Thanks @wbxl2000! - Fix new users getting stranded on "Model setup required" with no way back to sign-in when the first login finishes authorization but fails to complete model setup; the screen now offers a path back to the sign-in page so login can be retried.

• #2402 0f3b106 Thanks @wbxl2000! - Reword the sign-in waiting message from "Waiting for authorization" to "Waiting for authentication".

• Updated dependencies [40172c7]:

  • @moonshot-ai/kimi-code-sdk@0.15.0

0.6.5

Patch Changes

• #1994 beeb964 Thanks @RealKai42! - Reduce webview streaming re-render churn: settled assistant messages no longer re-render on every streaming delta, and local images over 10MB are no longer inlined into the webview DOM.
• Updated dependencies [ec88d35, b5efba7, ce0e3ce, e458323]:
  • @moonshot-ai/kimi-code-sdk@0.14.0

0.6.4

Changed

• Picking a model's highest thinking effort now applies to the current session only instead of becoming the global default: the top tier saves just the on/off toggle, lower tiers persist as the default as before, and re-confirming the current effort no longer rewrites the saved preference. The model and thinking pickers also note that switching mid-conversation invalidates the existing prompt cache.
• Unified the YOLO and Auto permission mode naming and descriptions with the CLI (/afk is now /auto), and approval requests that fall outside the active permission mode (sensitive files, plan reviews, ask rules) are now always shown to you instead of being auto-approved.

0.6.3

Fixed

• Editor mentions now work for files outside the working directory, and paths containing spaces are quoted correctly.
• Cancelling a running turn now reliably reaches the engine, and the UI no longer reports a task as stopped when there is nothing to cancel.
• Attaching to or resuming an existing session no longer overwrites its model and thinking effort with the configured defaults; model or effort changes picked in the composer are applied when the prompt is sent.

0.6.2

Fixed

• A core error arriving in the middle of a turn no longer corrupts the active turn; the turn now ends cleanly with an error instead of leaving the chat in a broken state.
• Kimi sign-in and connection failures now include the underlying transport cause (for example DNS or connection refused) instead of a generic error.
• Closed several FetchURL SSRF bypasses and the DNS-rebinding window.
• Tool calls interrupted mid-stream are now recorded and closed, so they no longer corrupt the session history.

0.6.1

Fixed

• The Sign in action in the settings (gear) menu now actually starts the Kimi login flow and shows an error toast when sign-in fails, instead of silently doing nothing.

0.6.0

Breaking

• Raised the minimum supported editor version to VS Code 1.100.0.
• Legacy Kimi Code OAuth credentials and MCP OAuth credentials are deliberately not migrated. Sign in to Kimi Code again and re-authorize affected MCP servers after upgrading.
• Removed the kimi.executablePath and kimi.environmentVariables settings. The old kimi.environmentVariables.KIMI_SHARE_DIR value is consulted only to discover legacy data during migration; it is not applied to the new runtime. The system-level KIMI_CODE_HOME environment variable remains supported.

Changed

• Replaced the legacy Python/stdio runtime with the in-process Kimi Code Node SDK. The extension no longer downloads or starts a separate Kimi executable.
• The in-process engine is the same one that powers the Kimi Code CLI, so the agent gains CLI-parity capabilities beyond the legacy runtime, including parallel subagent swarms, background tasks, and long-running goal runs.
• Added an opt-in legacy migration prompt on the first launch that detects data from version 0.5.x. The migration copies or merges supported data into the current Kimi Code home and does not delete the legacy source. If migration is skipped or needs to be retried, run Kimi Code: Migrate Legacy Data from the Command Palette.
• When VS Code and the Kimi Code terminal app resolve to the same KIMI_CODE_HOME, they use the same configuration and session storage. Running the same session concurrently from multiple processes is not supported or protected by cross-process locking.
• The model picker groups models by provider when multiple providers are configured, keeps provider identity when display names match, and recognizes adaptive-thinking metadata. A configured custom default provider no longer requires dismissing the Kimi account login screen on every launch.
• The file changes panel and Undo actions use extension-maintained baselines. Files changed through Kimi's Write and Edit operations are tracked on a best-effort basis. File deletions performed inside Bash are not tracked by this baseline and therefore cannot be restored by the panel's Undo action.

Fixed

• The kimi.yoloMode setting now reaches the permission engine: enabling it maps to the core yolo permission mode and takes effect when a session attaches, including sessions that previously stored a disabled auto-approve state.
• Kept the chat header and input toolbar readable when the sidebar is narrow: controls wrap and shrink instead of being clipped.

Distribution boundary

Release packaging produces target-specific VSIX files for darwin-x64, darwin-arm64, linux-x64, linux-arm64, win32-x64, and win32-arm64. Archive and static verification for a target does not by itself prove that the extension has run successfully in that target's Extension Host; runtime test results must be recorded separately for each operating system and architecture.