Kilo Release Notes

Minor Changes

• #9310 f067a90 - Support selecting Mercury Edit by Inception for autocomplete.
• #9548 c5614cc - Add Settings header buttons to open the project and global Kilo config files directly in VS Code.

Patch Changes

• #9552 3f0bf32 - Fix clearing an agent's Model Override in Agent Behaviour settings. Previously, clearing the field and saving would repopulate the old value because the empty input was sent as undefined and dropped by JSON.stringify, so the backend never received a delete instruction. The field now reverts to the global default model as expected.
• #9551 b344ac9 - Restore disabled provider management in the VS Code extension provider settings.

Minor Changes

• #9268 48c0553 - Open xterm.js-powered terminal tabs in the Agent Manager. Click the chevron next to the + tab button and pick "New Terminal" (or press Cmd+Shift+T / Ctrl+Shift+T) to spawn a real shell in the selected worktree or Local directory. Terminals render as proper tabs alongside agent sessions, support mixed drag-reorder with session tabs, and persist their position across webview reloads. The existing VS Code integrated terminal shortcut (Cmd+/) is unchanged.
• #9336 85c578e - Add the initial JetBrains session chat UI and improve sandbox debug logging for tracing chat events across frontend and backend.

Patch Changes

• #9335 6015ac6 - Restore explicit Submit behavior for single-choice question prompts in the VS Code extension so option clicks stay visible for review instead of immediately sending the answer.
• #9332 0bda9d1 - Fix mid-turn message handling so a new prompt sent while the assistant is working no longer aborts the in-flight response. The current LLM reply streams to completion, any pending suggestion or question is automatically dismissed, and the new prompt runs immediately after the current step instead of waiting for the entire multi-step turn to finish.
• #9119 8e75084 - Fix TUI freeze on huge-file diffs. Session-summary and file-view patches now use git directly instead of a JavaScript Myers implementation, so files of any size render a full diff without blocking the session.
• #9341 00ec003 - Significantly speed up LLM token streaming in long sessions. The chat view now stays responsive while the model streams a reply, even in sessions with hundreds of messages. Previously, each SSE batch produced ~1.3 seconds of visible freeze (roughly 80 dropped frames); streaming ticks are now inside a single animation frame.

Updated dependencies [00ec003]

• @opencode-ai/[email protected]
• @kilocode/[email protected]

Minor Changes

• #9207 4d85c23 - Support forking Agent Manager sessions from a specific user message.
• #9244 18442f6 - Support forking sidebar conversations from any user message.

Patch Changes

• #9305 b4cb8c3 - Prewarm the autocomplete backend on activation so inline completions are ready on first keystroke without opening the Kilo sidebar, and refresh autocomplete state when workspace folders change.
• #9318 bfe4482 - Prevent canceled autocomplete debounce requests from staying pending indefinitely.
• #9316 bf98bbe - Cap autocomplete ignore path caching so long sessions do not retain unlimited paths.
• #9239 2b17a7b - Fix custom provider model and variant deletions being silently reverted on save. Removing a model or reasoning variant from a custom provider now actually removes it from your config.
• #9241 6c0439b - Fix custom provider variant name validation showing inline errors correctly when a variant name is left empty
• #9208 acd5798 - Fixed Agent Manager incorrectly duplicating worktree sessions into the Local tab when forking a session or using "Continue in Worktree".
• #7960 d1aed36 - Add KiloClaw chat panel to the VS Code extension
• #9245 ec5ac2e - Fix the "Continue here" button not submitting after a plan is finished. Picking an option on a single-question prompt now sends the reply immediately — matching the CLI behaviour — and the redundant "Type your own answer" row no longer appears on the plan follow-up question.
• #9246 c529142 - Translate the follow-up prompt shown after finishing a plan. The "Ready to implement?" question, "Start new session" and "Continue here" buttons, and their descriptions now respect the language configured in the VS Code sidebar.
• #9129 60709fd - Fix duplicate text shown when typing a custom answer in the question tool. The preview text above the input field is now hidden while editing, so only the input value is visible.
• #9195 7038ce0 - Fix queued follow-up messages so they stay stacked at the bottom of the chat while the previous turn is still being processed, instead of getting interleaved with tool output.
• #9194 0ace2f9 - Restore earlier chat history when scrolling through virtualized message lists.
• #9236 6f17bb7 - Fix chat scroll jumping back to the bottom while the session is busy — you can now scroll up to read earlier context while a response is streaming or while a question is waiting for your answer.
• #9199 7a1388d - Fix a stuck session state when the suggest tool was left open. If the suggestion was never accepted or dismissed (for example, because VS Code was closed while it was showing), the session stayed marked as busy and any follow-up messages appeared queued forever. The session is now marked idle while waiting for a response to a suggestion.
• #9141 61ec925 - Restore spacing between sub-agent output and the following user message in the VS Code chat.

Minor Changes

• #8976 4ef6bbf - Support browsing and resuming sessions from Agent Manager with /sessions.
• #9023 5301258 - Support mentioning folders in the prompt with @ references, including top-level folder file contents.

Patch Changes

• #9121 c8fd421 - Fix the sidebar "Show Changes" diff viewer: the file tree now renders correctly (previously the file rows were cramped onto a single line due to missing styles), and per-file revert buttons are available, matching the Agent Manager.
• #9046 671129d - Fix a native memory leak on Windows where kilo serve would grow to several GB of RAM within minutes of opening the Agent Manager. Git diff polling now runs directly in the extension host instead of routing through the CLI subprocess, and the diff detail view caps per-file reads at 20 MB to prevent memory spikes when opening very large files.
• #9118 343455b - Respect per-agent model selections when an agent has a model configured in kilo.jsonc. Switching the model for such an agent now sticks across agent switches and CLI restarts. To pick up a newly edited agent default, re-select the model once (or clear ~/.local/share/kilo/storage/model.json).
• #9067 959a8b4 - Fix "assistant prefill" errors when a user queues a prompt while the previous turn is still streaming. The queued message no longer lands in the middle of the prior turn's history, so the next request always ends with the user prompt.
• #9123 9749cc1 - Show MCP tool call inputs alongside outputs in chat, with JSON syntax highlighting for both.
• #8911 eac2dba - Make switching between sessions in Agent Manager near-instant. Long sessions no longer freeze the UI when selected, and the chat view self-heals if it missed any messages while the session was in the background.
• f270639 - Show an inline error in the Settings save bar when the configuration fails to save (for example, due to an invalid value) so the user can correct the config and retry instead of losing their unsaved changes silently.

Updated dependencies [eac2dba]

• @opencode-ai/[email protected]
• @kilocode/[email protected]

Minor Changes

#9099 49b283e - Add a "Contribute on GitHub" call-to-action in the Marketplace view so users can easily propose a new skill, mode, or MCP server. The CTA appears as a subtle footer below the card grid and inside the empty search-results state, linking to the kilo-marketplace repository.

Patch Changes

#9066 79ba643 - Add xhigh reasoning effort option to custom provider model variants so users can access the highest tier on models that support it (gpt-5.2, gpt-5.3, gpt-5.4, gpt-5.1-codex-max, etc.).

#9068 e65c2d9 Thanks @kilo-code-bot! - Hide Kilo Gateway models that do not support tool calling from the model list.

#9040 7d2f5a7 - Preserve local Agent Manager sessions after the panel reloads while still pruning stale worktree sessions.

#9057 1526a4b - Keep the active chat session responsive while other sessions stream in the background. Applies to both the sidebar and the Agent Manager.

Preserve write tool alongside apply_patch for GPT-5 models (@jacksonkasi1)

• Add kilo help --all command for full CLI reference in markdown or text (@maphew)
• Auto-generate CLI reference docs from help.ts (@maphew)
• Improve config warning UX across CLI, TUI, and VS Code
• Surface schema validation errors to user in CLI
• Skip invalid agent/command configs instead of crashing
• Align /local-review diff with agent manager diff viewer
• Various CLI documentation improvements
• Thanks also to @justincqz, @thomasboom, @ScuttleBot

Preserve write tool alongside apply_patch for GPT-5 models (@jacksonkasi1)

• Add kilo help --all command for full CLI reference in markdown or text (@maphew)
• Auto-generate CLI reference docs from help.ts (@maphew)
• Improve config warning UX across CLI, TUI, and VS Code
• Surface schema validation errors to user in CLI
• Skip invalid agent/command configs instead of crashing
• Align /local-review diff with agent manager diff viewer
• Various CLI documentation improvements
• Thanks also to @justincqz, @thomasboom, @ScuttleBot

Preserve write tool alongside apply_patch for GPT-5 models (@jacksonkasi1)

• Add kilo help --all command for full CLI reference in markdown or text (@maphew)
• Auto-generate CLI reference docs from help.ts (@maphew)
• Improve config warning UX across CLI, TUI, and VS Code
• Surface schema validation errors to user in CLI
• Skip invalid agent/command configs instead of crashing
• Align /local-review diff with agent manager diff viewer
• Various CLI documentation improvements
• Thanks also to @justincqz, @thomasboom, @ScuttleBot

Core

• Preserve write tool alongside apply_patch for GPT-5 models (@jacksonkasi1)
• Strengthen sync test and add missing ConfigCLICommand
• Align /local-review diff with agent manager diff viewer
• Make 'kilo help' show same output as 'kilo -h'
• Restore native --help for subcommands
• Add kilocode markers on mode block
• Add kilocode markers on agent block
• Add kilocode markers on command block
• Add workflows section to CLI documentation
• Add KILO_CONFIG_DIR to command paths in CLI documentation
• Add sync check ensuring commands.ts and index.ts stay in sync
• Disable built-in help command to allow kilo help --all to work
• Track remote connection opened event in PostHog
• Drop explicit CommandModule type from commands array to fix typecheck
• Add completion command to commands array for help --all and cli-reference
• Minimize index.ts diff and remove spurious changes
• Guard session import in config error handlers
• Fix kilocode_change markers in CLI code
• Let managed config errors propagate
• Include detail in validation error
• Move commands barrel to kilocode directory and restore upstream-style index.ts chain
• Improve config warning UX across CLI, TUI, and VS Code
• Add documentation explaining the purpose of commands.ts barrel file
• Add RemoteCommand to barrel exports, fix recursive subcommands, and include $0 commands in help --all output
• Add config warnings endpoint and surface errors to TUI/extension
• Add CONTEXT.md to CLI instructions list
• Add KILO_CONFIG_DIR to command lookup
• Widen KILO_DISABLE_PROJECT_CONFIG description in CLI documentation
• List all managed configuration files in CLI documentation
• Fix instructions auto-discovery list in CLI documentation
• Add missing opencode.jsonc to global paths documentation
• Add named-command lookup guidance to kilo-config skill
• Widened kilo-config skill to cover path location questions
• Add legacy paths to kilo-config skill documentation
• Surface schema validation errors to user in CLI
• Remove unscoped session.error publish from CLI
• Skip invalid agent/command configs instead of crashing
• Add regression tests for config resilience
• Remove empty openapi.json artifact and add DbCommand to test fixture (@maphew)
• Address bot review concerns (@maphew)
• Remove invalid ResolveMessage import and fix cli-reference link path (@maphew)
• Skip undescribed commands in review process and rename ambiguous variable (@maphew)
• Address review feedback: await rejects, gate options.all, sanitize cwd, add AttachStub (@maphew)
• Auto-generate CLI reference docs from help.ts (@maphew)
• Log catch block and document yargs version dependency (@maphew)
• Add kilo help --all command for full CLI reference in markdown or text (@maphew)

TUI

• Add kilocode_change marker to prompt flexShrink prop
• Add kilocode_change markers in toast code
• Add flexShrink={0} to prompt border box to prevent shrinking

Thank you to 6 community contributors:

@maphew:

• feat: add kilo help --all command for full CLI reference in markdown or text
• the plan for implementing help --all cmd
• fix: address review — log catch block, document yargs version dependency
• feat: auto-generate CLI reference docs from help.ts
• fix: address review — await rejects, gate options.all, sanitize cwd, add AttachStub
• fix: address review — skip undescribed commands, rename ambiguous var, remove marker
• Merge remote-tracking branch 'origin/main' into feat/help-all-command
• Merge upstream/main into feat/help-all-command; add DbCommand and WorkspaceServeCommand
• fix: remove invalid ResolveMessage import and fix cli-reference link path
• Merge branch 'main' into feat/help-all-command
• fix: address bot review concerns
• fix: remove empty openapi.json artifact, add DbCommand to test fixture

@aravhawk:

• Merge branch 'Kilo-Org:main' into session/agent_58776529-50a0-40ff-9d78-3f2f7e4d28b6

@justincqz:

• remove references to 'kilo web' in documentation

@thomasboom:

• Update bonus credits and AI model versions in README
• Merge branch 'main' into docs/updated-models-mentioned

@jacksonkasi1:

• fix: preserve write tool alongside apply_patch for gpt-5 models

@ScuttleBot:

• docs: update for task timeline graph and Agent Manager PR badges (#8532)

Core

• Guard against prompt injection in commits
• Add scope context and better git commands to review prompt
• Inject --rm flag for Docker MCP containers to prevent accumulation (@johnnyeric)
• Comment out specific conditions in ID check logic
• Add support for GLM, Kimi, and Qwen reasoning models
• Follow-up execution is now aware of the saved plan file (@shssoichiro)
• Update minimatch, @modelcontextprotocol/sdk, and @aws-sdk dependencies
• Update Hono to fix authentication bypass and server vulnerabilities
• Update simple-git dependency to fix critical remote code execution vulnerability
• Preserve specific MCP tool rules when propagating permissions to sub-agents
• Propagate MCP restrictions to sub-agents alongside edit and bash
• Preserve inherited restrictions across multi-hop sub-agent chains
• Apply read-only bash and MCP restrictions to plan mode and propagate bash restrictions to sub-agents
• Plan mode now respects edit restrictions and sub-agents inherit caller's file access permissions

Thank you to 2 community contributors:

• @shssoichiro:
  fix(core): make follow-up execution aware of the saved plan file
• @johnnyeric:
  fix(mcp): inject --rm flag for Docker MCP containers to prevent accumulation