43 KiB
Feature Parity Plan — Kilo Code VS Code Extension (Rebuild)
Overview
This extension is a ground-up rebuild of the old Kilo Code extension using Kilo CLI as the backend. Rather than migrating the old extension's codebase, we started fresh with a Solid.js webview, a CLI server manager, and a message-based protocol between extension host and webview. This new extension lives in the Kilo monorepo.
This document tracks remaining work needed for feature parity with the old extension. Each feature links to its detailed parity requirement doc. Features sourced from the GitHub project board include issue links.
Current State Summary
The rebuild has a working foundation:
- CLI backend: server lifecycle (spawn, port detection, auth, dispose) in
server-manager.ts, HTTP client with 20 endpoints inhttp-client.ts, SSE client with event subscriptions insse-client.ts - Chat UI: message list with text/tool/reasoning parts, streaming text deltas, auto-scroll in
ChatView.tsxandMessageList.tsx. Markdown rendering via kilo-ui's<KiloMessage>component with syntax-highlighted code blocks and per-tool renderers. - Tool parts: delegation to kilo-ui per-tool renderers with status icons, expandable input/output sections, status-based styling in
Message.tsx - Prompt input: send/abort controls with integrated model selector and mode switcher in
PromptInput.tsx - Permissions: rendered through kilo-ui's DataProvider/permission components — reject/once/always with expandable tool details
- Sessions: create, list, select, load messages via
session.tsx - Todo list:
todo.updatedSSE event → session store → kilo-uiPART_MAPPING["todowrite"]renders todo lists with checkboxes - Auth: full device auth flow with QR code, verification code, countdown in
DeviceAuthCard.tsx; org switching implemented - Profile: login state, balance, dashboard link, logout in
ProfileView.tsx - Session history: search, date-grouped list, relative timestamps, rename/delete in
SessionList.tsx - Settings: 14-tab settings (Providers, AgentBehaviour, AutoApprove, Browser, Autocomplete, Display, Notifications, Context, Terminal, Prompts, Experimental, Language, AboutKiloCode — all functional; Workflows subtab is a placeholder) in
Settings.tsx - Message protocol: 41 message types (22 ExtensionMessage + 19 WebviewMessage) in
messages.ts - Build pipeline: dual esbuild (extension + webview), CLI binary provisioning in
esbuild.jsandprepare-cli-binary.mjs - View title bar button: ✅ Done — #181
- Browser automation: Playwright MCP integration with settings toggle, lifecycle service, and CLI MCP hub registration in
BrowserAutomationService - kilo-ui integration: Webview uses
@kilocode/kilo-uishared component library — ThemeProvider, MarkedProvider, DataProvider, Button, IconButton, Tooltip, Popover, Switch, Select, Toast, KiloMessage - Task header: Session title, cost display, context token usage with percentage, compact button in
TaskHeader.tsx - Model selector: Popover-based model picker with search, provider grouping, free model tags in
ModelSelector.tsx - Mode switcher: Agent/mode selector with descriptions in
ModeSwitcher.tsx - Localization: Full i18n system with 16 locales, three-layer dict merging, locale auto-detection in
language.tsxandi18n/ - Context compaction: summarize endpoint + CompactRequest message + compact button in TaskHeader
- Code blocks: copy buttons, expand/collapse, syntax highlighting via kilo-ui KiloMessage + MarkedProvider + shiki
- Diff viewing: full diff viewer via kilo-ui
Diff/DiffChangescomponents for file edit/write/patch tools - Command execution: expandable bash tool output in BasicTool accordion via kilo-ui PART_MAPPING["bash"]
- Image handling: clipboard paste, drag-and-drop, thumbnails in PromptInput, ImagePreview dialog in
useImageAttachments.ts - Autocomplete: FIM inline completions via
AutocompleteInlineCompletionProviderwith debouncing, LRU cache, tree-sitter context - Agent Manager: parallel worktree sessions, git worktree lifecycle, WorktreeSelector in
AgentManagerProvider.ts - Telemetry:
TelemetryProxyPOSTs to CLI/telemetry/capture; webview telemetry forwarded via message protocol - Auto-approval config: per-tool allow/ask/deny dropdowns for all 16 tools in AutoApproveTab
- Provider configuration: default/small model selection, disabled/enabled provider lists in ProvidersTab
Chat UI Feature Parity
| Feature | Status | Details | Backend | Priority |
|---|---|---|---|---|
| Auto-Approval Controls | ✅ Done | AutoApproveTab in Settings implements per-tool allow/ask/deny dropdowns for all 16 tools with a "Set All" bulk control. Changes written to backend config via updateConfig({ permission: {...} }). #171 |
CLI owns permissions; webview needs config UI | P1 |
| Browser Session Controls | 🔨 Partial | BrowserTab settings UI implemented with enable/disable toggle, system Chrome and headless options. Missing: in-chat browser session controls, action replay, screenshot viewing. | CLI-side (if browser tool exists) + webview | P3 |
| Checkpoint & Task Management | ❌ Not started | No checkpoint restore, navigation, or "See New Changes" diff buttons. | CLI session undo/redo/fork + extension git integration | P1 |
| Code Block Interactions | ✅ Done | Code blocks render with syntax highlighting, copy buttons, and expand/collapse via kilo-ui's KiloMessage + MarkedProvider + shiki. User messages and assistant text parts both have copy buttons. | Webview-only | P0 |
| Command Execution | ✅ Done | Bash tool output renders in an expandable BasicTool accordion via kilo-ui's PART_MAPPING["bash"]. Command output shown in a scrollable code block with ANSI stripping. | CLI executes commands; webview renders output | P0 |
| Connection State UI | 🔨 Partial | ServerProvider context tracks connectionState ("connecting"/"connected"/"disconnected"/"error"). Prompt input is disabled when disconnected. Missing: dedicated blocking spinner/error panel overlay. |
Webview-only (consumes connection state) | P0 |
| Context Menus & Tooltips | ❌ Not started | No right-click context menus or hover tooltips on interactive elements. | Webview-only | P2 |
| Diff Viewing & File Operations | ✅ Done | Full diff viewer via kilo-ui Diff + DiffChanges components (built on @pierre/diffs) with line selection, find shortcuts, and worker-pool syntax highlighting. DiffComponentProvider registered in App.tsx. File edit/write/patch tools all use this. |
CLI provides diff data; webview renders | P0 |
| File Permission Dialogs | 🔨 Partial | Basic permission dialog exists (reject/once/always). Missing batch file read approval and per-file granularity. #171 | CLI permission model; webview UI | P1 |
| Follow-Up Questions | ❌ Not started | No suggested reply chips, click-to-submit, auto-approval countdown, or mode indicators. | Likely extension-side generation | P2 |
| Image Handling | ✅ Done | useImageAttachments hook handles clipboard paste, drag-and-drop, and file reading. Thumbnails with remove buttons shown in PromptInput. kilo-ui ImagePreview dialog for viewing attached images in messages. |
CLI provides image data; webview renders + VS Code integration | P1 |
| Inline Actions on Tool Messages | 🔨 Partial | Tool parts render with status icons (⏳⚙️✓✕), expandable sections with input/output, and status-based CSS classes. File path links render as clickable anchors in tool subtitles via kilo-ui. Missing: explicit VS Code file-opener wired via postMessage, inline action buttons. | CLI provides tool metadata; webview renders | P1 |
| Kilo Themed Chat Session | ✅ Done | kilo-ui components throughout: ThemeProvider with defaultTheme="kilo-vscode", KiloMessage for rendering, MarkedProvider for markdown, Button/IconButton/Tooltip/Popover from kilo-ui. #161 |
Webview-only (styling/theming) | P0 |
| Markdown Rendering | ✅ Done | Message.tsx delegates to kilo-ui's <KiloMessage> component. MarkedProvider in App.tsx provides markdown context. Syntax highlighting via shiki. #161 |
Webview-only | P0 |
| Mermaid Diagram Features | ❌ Not started | No mermaid rendering, "Fix with AI" button, copy, or open-as-PNG. Requires markdown rendering first. | Webview-only (rendering); CLI for "Fix with AI" | P2 |
| Message Editing & Management | ❌ Not started | No inline editing, deletion, timestamp display, or redo-previous-message (up-arrow). #177 | CLI session fork/undo for edit semantics | P1 |
| Special Content Types | 🔨 Partial | Reasoning blocks render (collapsible). Missing: open-markdown-preview button, MCP tool/resource rows, expandable error rows with copy. | Mixed: CLI for MCP data; webview for rendering | P1 |
| Task Header | ✅ Done | TaskHeader.tsx implements session title, cost display (formatted USD), context token usage with percentage, compact button with tooltip. Uses kilo-ui IconButton and Tooltip. #166 | CLI provides cost/context data; webview renders | P0 |
| Todo List Management | ✅ Done | todo.updated SSE event handled through full pipeline: KiloProvider → webview message → session store. kilo-ui PART_MAPPING["todowrite"] renders todo lists with checkboxes showing completed/total counts. |
CLI tool or extension-side feature | P2 |
| Toggle Thinking | 🔨 Partial | Reasoning blocks render but no toggle to enable/disable thinking. Linked PR #127. #172 | CLI controls thinking; webview provides toggle UI | P2 |
Non-Agent Feature Parity
| Feature | Status | Details | Backend | Priority |
|---|---|---|---|---|
| Agent Manager | ✅ Done | AgentManagerProvider opens editor panel, creates git worktrees for parallel sessions, full createWorktreeSession lifecycle (worktree → session → first message). WorktreeManager handles git worktree add, metadata, and discovery. WorktreeSelector in PromptInput toggles local/worktree mode. #178 |
Extension orchestrates multiple CLI sessions | P1 |
| Authentication & Enterprise | 🔨 Partial | Device auth flow works, organization switching implemented. Missing: org feature flags, MDM policy enforcement. #160 | CLI handles its auth; extension handles org/MDM | P1 |
| Auto-Purge | ❌ Not started | No scheduled cleanup of old session/task storage. | Extension-side (storage ownership TBD) | P3 |
| Autocomplete / Ghost | ✅ Done | AutocompleteInlineCompletionProvider implements vscode.InlineCompletionItemProvider with FIM completions via Kilo Gateway. Includes debouncing, LRU cache, tree-sitter context, bracket matching, and stream-based completion. AutocompleteServiceManager registers/unregisters the provider. #164 |
Extension-side (VS Code InlineCompletionProvider) | P1 |
| Browser Automation & URL Ingestion | 🔨 Partial | Playwright MCP integration implemented: settings toggle, BrowserAutomationService, CLI MCP hub registration, BrowserTab settings UI. Missing: URL-to-markdown ingestion, screenshot viewing in chat. | CLI MCP hub (POST /mcp); extension manages lifecycle | P3 |
| Checkpoints | ❌ Not started | No shadow git repo, per-task snapshots, restore UI, or diff viewing. Settings tab is a stub. | CLI (partial: session undo/redo); extension for git snapshots | P1 |
| Cloud Task Support | ❌ Not started | No cloud sync for tasks across devices. #168 | Kilo cloud API + CLI; extension provides UI | P2 |
| Code Actions & Editor Menus | 🔨 Partial | Editor/terminal right-click submenus, CodeActionProvider lightbulb, keyboard shortcuts, and prompt templates implemented. Terminal content capture is a stub (needs shell integration API). Missing: custom prompt overrides via settings. | Extension-side (VS Code CodeActionProvider + menus + keybindings) | P1 |
| Code Reviews | ❌ Not started | No local review mode or automated AI review of uncommitted/branch changes. | CLI (partial); extension for VS Code review UX | P2 |
| Codebase Indexing & Semantic Search | ❌ Not started | No vector indexing, semantic search, or embeddings infrastructure. | CLI has grep/glob endpoints; semantic indexing is extension or cloud | P2 |
| Contribution Tracking | ❌ Not started | No AI attribution tracking, line fingerprinting, or reporting. | Extension-side | P3 |
| Custom Commands | ❌ Not started | No slash commands, project-level command discovery, or YAML frontmatter support. | CLI has custom commands; extension provides UI entry points | P2 |
| Deploy & Secure Surfaces | ❌ Not started | No deploy workflows, managed indexing UI, or security review surfaces. | Extension-side | P3 |
| Fast Edits | ❌ Not started | No fast edit mode for quick inline code changes. | CLI fast-edit runtime; extension provides UI | P2 |
| Git Commit Message Generation | ❌ Not started | No AI commit message generation or VS Code Source Control integration. #165 | Extension-side (VS Code Git API) | P2 |
| Integrations | ❌ Not started | No external system integrations (GitHub, etc.) beyond basic auth. | CLI plugin system (partial); extension for IDE hooks | P3 |
| Kilo Gateway | ✅ Done | Kilo Gateway (KILO_GATEWAY_ID = "kilo") is the default provider, listed first in ModelSelector via PROVIDER_ORDER. Login/logout/org switching fully implemented in KiloProvider.ts and ProfileView.tsx. Default model selection in ProvidersTab. #176 |
CLI handles gateway connection; extension provides config UI | P0 |
| Localization | ✅ Done | Full i18n system with 16 locales in i18n/, three-layer dict merging, locale auto-detection, LanguageTab settings UI with locale selector. |
Extension + webview; CLI locale mapping needed | P3 |
| Marketplace | 🔨 Partial | Placeholder view exists but is non-functional. No catalog, install, or update capabilities. #169 | Extension-side | P2 |
| MCP & MCP Hub | 🔨 Partial | MCP types and HTTP client methods added (getMcpStatus, addMcpServer, connectMcpServer, disconnectMcpServer). Used by BrowserAutomationService. Missing: general MCP configuration UI, server management, tool allowlisting, connection status display. | CLI owns MCP lifecycle; extension provides config UI | P1 |
| Mode Switcher | ✅ Done | ModeSwitcher.tsx implements popover-based agent/mode selector with descriptions, persisted via session.selectAgent(). Integrated into PromptInput. #162 | CLI manages modes; extension provides switcher UI | P2 |
| Model Switcher | ✅ Done | ModelSelector.tsx implements popover-based selector with search/filter, keyboard navigation, provider grouping, free model tags. Integrated into PromptInput. #163 | CLI provides model list; extension provides switcher UI | P1 |
| Provider Configuration | 🔨 Partial | Provider context fetches and exposes provider data, connected providers, defaults, model lists — wired into ModelSelector. ProvidersTab UI still a stub. #175 | CLI manages providers; extension provides config UI | P1 |
| Repository Initialization | ❌ Not started | No /init command support for setting up agentic engineering. #174 | CLI /init endpoint; extension provides UI trigger | P3 |
| Rules & Workflows | ❌ Not started | No rules or workflow management UI. #173 | CLI owns rules runtime; extension provides management UI | P3 |
| Search & Repo Scanning | ❌ Not started | No search infrastructure beyond CLI grep/glob. | CLI has grep/glob; extension may add UI | P2 |
| Settings Sync | ❌ Not started | No VS Code Settings Sync allowlist registration. Settings tabs are all stubs. | Extension-side (VS Code API) | P3 |
| Settings UI | 🔨 Partial | 15-tab settings shell exists. BrowserTab has real settings controls (enable/disable, system Chrome, headless toggles). LanguageTab has working locale selector. Remaining 13 tabs are stubs. #170 | CLI exposes config; extension provides settings forms | P1 |
| Skills System | ❌ Not started | No skill discovery, management, or hot-reload in extension. | CLI has skills runtime; extension provides packaging/UI | P2 |
| Speech-to-Text | ❌ Not started | No voice input or streaming STT. | Webview (mic capture); CLI-compatible STT optional | P3 |
| Task History | 🔨 Partial | Session list exists but lacks search, metadata, and full persistence. #167 | CLI session storage; extension provides history UI | P1 |
| Telemetry | ❌ Not started | Dual-layer telemetry (PostHog Node server-side + PostHog JS client-side) for extension usage, errors, LLM completions, and AI interactions. Includes privacy controls, typed events, and structured error tracking. See detailed plan. | Extension-side (PostHog + kilo-telemetry) | P1 |
| Terminal / Shell Integration | ❌ Not started | No VS Code terminal integration for command execution display, exit code tracking, or working directory changes. | CLI executes commands; extension provides terminal UX | P1 |
Infrastructure & Robustness
These items were identified from the JetBrains plugin analysis — patterns the JetBrains plugin implements that are missing in the VSCode extension. They primarily affect reliability and developer experience rather than feature parity.
| Feature | Status | Details | Scope | Priority |
|---|---|---|---|---|
| SSE Auto-Reconnect | ❌ Not started | SSE connection has no reconnect logic. Network hiccups, laptop sleep, or server restart kill the extension with no recovery. | Extension (SSEClient + ConnectionService) | P0 |
| HTTP Request Timeouts | ❌ Not started | HTTP client uses bare fetch() with no timeouts. Hung requests block indefinitely. |
Extension (HttpClient) | P1 |
| VSCode Error Notifications | ❌ Not started | Critical errors (CLI missing, server crash) are only shown in the webview. No vscode.window.showErrorMessage() notifications. |
Extension (KiloProvider) | P1 |
| Dedicated Output Channel | ❌ Not started | All logging goes to console.log mixed with other extensions. No dedicated "Kilo Code" output channel. | Extension (new logger utility) | P2 |
Pre-Production Checklist
Before publishing this extension to the VS Code Marketplace or deploying to users, verify every item below.
Security
- Review and tighten CSP — The current policy in
KiloProvider._getHtmlForWebview()has several areas to audit:style-src 'unsafe-inline'is broadly permissive — investigate whether nonce-based style loading is feasible now that kilo-ui styles are bundledconnect-src http://127.0.0.1:* http://localhost:*allows connections to any localhost port — tighten to the actual CLI server port once known at runtimeimg-src … https:allows images from any HTTPS origin — scope to${webview.cspSource} data:unless external images are explicitly needed'wasm-unsafe-eval'inscript-srcwas added for shiki — confirm it is still required and document the reasonws://connections to any localhost port — same concern asconnect-src
- Validate
openExternalURLs — TheopenExternalhandler passes any URL from the webview directly tovscode.env.openExternal()with no allowlist or scheme check. Restrict tohttps:(and possiblyvscode:) schemes, or allowlist specific hosts - Audit credential storage — CLI stores credentials as plaintext JSON with
chmod 0600. Evaluate whether VS Code'sSecretStorageAPI should be used for extension-side secrets, and document the threat model for CLI-managed credentials - Audit workspace path containment — CLI's path traversal checks are lexical only; symlinks and Windows cross-drive paths can escape the workspace boundary. Determine if additional hardening (realpath canonicalization) is needed before production
Reliability
- VS Code error notifications — Critical errors (CLI missing, server crash, connection lost) are only shown inside the webview (details). Users get no feedback if the webview is hidden
- Connection state UI — No loading spinner, error panel, or reconnecting indicator in the webview (details). Chat renders even when disconnected
Testing
- Test coverage — Only one test file exists (
extension.test.ts). Add integration tests for: server lifecycle, SSE event routing, message send/receive, permission flow, session management - Multi-theme visual check — Verify the webview renders correctly in at least one light theme, one dark theme, and one high-contrast theme
- Multi-platform smoke test — Test on macOS, Windows, and Linux. Particularly: CLI binary provisioning, path handling,
chmod-based credential protection on Windows
Packaging & Marketplace
- Bundle size audit — With kilo-ui and its transitive dependencies (shiki, marked, katex, dompurify, etc.) now bundled, measure
dist/webview.jssize and verify the total.vsixpackage size is acceptable .vscodeignorereview — Ensure only necessary files are included in the package (nodocs/,src/, test artifacts, or development scripts)- Marketplace metadata — Verify
README.md,CHANGELOG.md, publisher name, extension icon, andpackage.jsonfields (displayName,description,categories,keywords,repository) are production-ready activationEventsreview — Confirm the extension only activates when needed (not*), to avoid impacting VS Code startup time- Minimum VS Code version — Verify
engines.vscodeinpackage.jsonmatches the minimum API features actually used
Logging & Observability
- Dedicated output channel — All logging currently goes to
console.logmixed with other extensions (details). Create a dedicated "Kilo Code" output channel before production - Remove or guard verbose logging — Many
console.logcalls with emojis and debug detail exist inKiloProvider.ts. Gate behind a debug flag or move to the output channel at appropriate log levels
Implementation Notes
Architecture
- Solid.js (not React) powers the webview. JSX compiles via
esbuild-plugin-solid. All webview components use Solid's reactive primitives (signals, createEffect, etc.). - Two separate esbuild builds: extension (Node/CJS) and webview (browser/IIFE), configured in
esbuild.js. - No shared state between extension and webview. All communication is via
vscode.Webview.postMessage()with typed messages defined inmessages.ts. Provider hierarchy:ThemeProvider → DialogProvider → VSCodeProvider → ServerProvider → LanguageBridge → MarkedProvider → ProviderProvider → SessionProvider → DataBridge. - CLI backend owns: agent orchestration, MCP lifecycle, tool execution, search/grep/glob, session storage, permissions runtime, custom commands, skills, and fast edits.
- Extension owns: VS Code API integrations (code actions, inline completions, terminal, SCM, settings sync), webview rendering, auth mediation, and any feature not supported by CLI.
kilo-ui Shared Library
- kilo-ui shared library: The webview now heavily uses
@kilocode/kilo-uifor UI components. ADataBridgecomponent in App.tsx adapts the session store to kilo-ui'sDataProviderexpected shape, enabling shared components like<KiloMessage>to work with the extension's data model.
Key Differences from Old Extension
- No
Task.tsorwebviewMessageHandler.ts— the CLI server replaces the old in-process agent loop. - Permissions flow through CLI's ask/reply model, not extension-side approval queues. Permissions are rendered through kilo-ui's DataProvider pattern, not a standalone PermissionDialog.
- Session history is CLI-managed, not stored in VS Code global state.
- MCP servers are configured and managed by the CLI, not the extension.