codeep 2.14.0 → 2.16.0

package/README.md

@@ -23,27 +23,14 @@
   <a href="https://github.com/VladoIvankovic/Codeep"><img src="https://img.shields.io/github/stars/VladoIvankovic/Codeep?style=social" alt="GitHub stars"></a>
 </p>
 
-## Upgrading from 1.x to 2.0
+## Upgrading
 
-Most users **do not need to do anything** — `npm install -g codeep@latest`
-picks up the new version and everything carries over (config, saved
-profiles, project memory, MCP servers, skill bundles).
-
-Two breaking changes only affect a small audience:
-
-1. **MCP `clientInfo` version bumped from `1.4.0` → `2.0.0`.** If you
-   maintain a custom MCP server that allowlists Codeep by version
-   string, update your check.
-2. **`McpServer` protocol shape** in ACP now has `command?`, `args?`,
-   `url?`, `headers?` (transport-agnostic for stdio + Streamable HTTP).
-   The old required-`command` shape is still accepted — parser handles
-   both — but new clients should emit the optional shape.
-
-See [CHANGELOG.md](CHANGELOG.md#200--2026-05-18) for the full list of
-additions (full MCP support, OpenRouter provider with authoritative
-per-call cost, Claude-Code-compatible skill bundles + web marketplace,
-custom slash commands, lifecycle hooks, checkpoints, `/cost`,
-`/compact`, and more).
+Running `npm install -g codeep@latest` picks up the new version — config,
+saved profiles, project memory, MCP servers, and skill bundles all carry
+over. Breaking changes (if any) are listed at the top of each release in
+[CHANGELOG.md](CHANGELOG.md); the last breaking change was
+[2.0.0](CHANGELOG.md#200--2026-05-18) (MCP `clientInfo` version bump,
+transport-agnostic `McpServer` shape).
 
 ## Features
 
@@ -77,6 +64,7 @@ When started in a project directory, Codeep automatically:
 - **AI session titles** - Sessions auto-title themselves with an LLM one-liner ("OAuth2 migration for auth module") instead of a truncated first message. This makes one small background API call per session; turn it off with the `autoSessionTitle` setting (`/settings`) if you prefer zero unsolicited calls
 - **Search current session** - Find text in the open conversation with `/search`
 - **Cross-session recall** - `/recall <query>` searches across **all** saved sessions, ranked by relevance + recency. Add `--resume` to load the top match, or `--summarize` for an LLM recap of what you accomplished across matches
+- **Cross-device resume (`/cloud`)** - Sessions you run on one device (CLI, Mac app, VS Code) sync to `codeep.dev`. `/cloud` lists them and pulls the selected one back locally, so you can start a session on your desktop and continue on your laptop. Run `codeep account` once to link your devices; scoped to the current project when one is open.
 - **Export** - Save to Markdown, JSON, or plain text
 - `/cost` - Per-session token usage and estimated cost (per provider/model)
 - `/compact [keepN]` - AI-summarize older messages to free up context (keeps last N, default 4)
@@ -101,15 +89,19 @@ When started in a project directory, Codeep automatically:
 - Press Enter to send, Escape to cancel
 - Works reliably in all terminals (no Ctrl+V issues)
 
-### File Context (`/add`, `/drop`)
+### File Context (`/add`, `/drop`, `@mentions`)
 Explicitly add files to the conversation context:
 
 - **`/add <path>`** - Add one or more files to context
 - **`/add`** (no args) - Show currently added files
 - **`/drop <path>`** - Remove a specific file from context
 - **`/drop`** (no args) - Remove all files from context
+- **`@path/to/file`** - Inline mention: type `@` followed by a file path anywhere in your message and its contents are attached for that message only. Supports relative paths (`@src/index.ts`), absolute (`@/etc/hosts`), home (`@~/.codeep/profile.md`), and quoted paths with spaces (`@"my file.ts"`). In the CLI, typing `@` opens an autocomplete file picker — ↑↓ to navigate, Tab to insert.
+- **`@folder <path>`** (alias `@dir`) - Attach an entire directory: type `@folder src/components` and every source file under it is loaded recursively. Skips `node_modules`, `.git`, `dist`, binary extensions (images, archives, lockfiles), and files over the per-file size cap. Stops at a 200 KB total cap per mention (with a notification when it hits the limit). Quoted paths supported: `@folder "my components"`.
+- **`@web <url>`** - Inline web fetch: type `@web https://example.com/docs` (or `@web example.com/docs`) and the page is fetched, converted to readable text, and attached to the message. HTML is stripped; plain text and JSON pass through as-is. Capped at 32 KB, 12 s timeout. Successful fetches are cached for 30 min (per session, up to 50 entries) so repeated `@web` for the same URL is instant and free; use `/web-cache clear` to reset.
+- **`@git <ref>`** - Inline git context: type `@git diff`, `@git diff --staged`, `@git HEAD`, `@git <sha>`, `@git main:src/x.ts`, or `@git diff a..b` to inject that diff / commit / file-at-ref into the message as a `[Git ref]` block. Capped at 64 KB per mention (truncated with a marker). Unknown refs surface a friendly failure note.
 
-Added files are automatically attached to every message (both chat and agent mode) until dropped. Useful for giving the AI specific files to work with.
+Added files (via `/add`) are automatically attached to every message (both chat and agent mode) until dropped. `@mentions` are per-message — handy for one-off references without polluting the persistent context.
 
 ```
 > /add src/utils/api.ts src/types/index.ts
@@ -118,6 +110,10 @@ Added 2 file(s) to context (2 total)
 > refactor the API client to use async/await
 # AI sees both files attached to your message
 
+> review @src/auth/login.ts for security issues
+Loaded 1 file(s) from @mentions
+# AI sees login.ts attached for this message only
+
 > /drop
 Dropped all 2 file(s) from context
 ```
@@ -358,10 +354,11 @@ You write safe, reversible DB migrations…
 **Guaranteed review:** enable **Agent Auto-Review** in `/settings` (`agentAutoReview`) and after any run that changes files, Codeep automatically delegates to the `reviewer` and appends its findings — a review stage that always happens. Off by default.
 
 ### Project Rules
-Define project-specific instructions that the AI always follows. Create a rules file in your project root:
+Define project-specific instructions that the AI always follows. Create a rules file in your project root (first non-empty file wins):
 
 - **`.codeep/rules.md`** - Primary location (inside `.codeep/` folder)
 - **`CODEEP.md`** - Alternative location (project root, similar to `CLAUDE.md` or `.cursorrules`)
+- **`AGENTS.md`** - Cross-tool standard (works with Claude Code, Cursor, Kilo Code — no duplication needed)
 
 The rules file content is automatically included in every system prompt — both chat and agent mode. Use it to define:
 
@@ -469,11 +466,12 @@ Full spec coverage in 2.0.0 — stdio and Streamable HTTP transports,
 tools / resources / prompts / sampling, capability negotiation with
 `roots`, auto-restart on crash, mid-run catalog refresh.
 
-**Three ways to wire MCP servers** (all merge; project > global > ACP wins last):
+**Four ways to wire MCP servers** (all merge; project > `.mcp.json` > global > ACP wins last):
 
 1. **Project config** — `.codeep/mcp_servers.json` (committed with the repo)
-2. **Global config** — `~/.codeep/mcp_servers.json` (per machine)
-3. **ACP client config** — Zed / Claude Desktop pass `mcpServers` on `session/new`
+2. **`.mcp.json`** — cross-tool standard at the workspace root (works with Claude Code, Cursor, Kilo Code — point everyone at the same file)
+3. **Global config** — `~/.codeep/mcp_servers.json` (per machine)
+4. **ACP client config** — Zed / Claude Desktop pass `mcpServers` on `session/new`
 
 Two transports per entry, mutually exclusive:
 
@@ -573,6 +571,26 @@ cloned repo's hooks are **not** run until you approve the workspace. Run
 own projects just need a one-time `/hooks trust`. Global `~/.codeep/hooks/` are
 never run for the same reason.
 
+#### Windows notes
+
+Lifecycle hooks are POSIX shell (`.sh`) scripts:
+
+- **macOS / Linux** — run directly (via the script's shebang). No setup.
+- **Windows** — a `.sh` needs a POSIX `sh`. Install [Git for Windows](https://git-scm.com/download/win)
+  (it ships `sh.exe`) and Codeep runs `.sh` hooks through it automatically — no
+  PATH tweaking needed if Git is installed to its default location. **Without a
+  POSIX shell**, `/hooks` and the welcome banner report an explicit
+  *"cannot run on this system"* state and hooks are skipped — they never block a
+  tool call or commit, and the rest of Codeep is unaffected. (`.ps1`/`.cmd` hooks
+  are not executed; keep hook logic in `.sh`.)
+
+Everything else is **Node-native and shell-independent on Windows**: project
+rules (`AGENTS.md` / `CODEEP.md` / `.codeep/rules.md`), custom slash commands
+(`.codeep/commands/*.md`), and skill bundles (`.codeep/skills/`) are Markdown;
+MCP config (`.mcp.json` / `.codeep/mcp_servers.json`) is JSON. MCP servers spawn
+via their configured `command` — `npx` resolves through `npx.cmd`, and a custom
+`command` like `python` just needs to be on your PATH.
+
 ### Skill Bundles (new in 2.0)
 Beyond the built-in skills and custom slash commands, Codeep now supports
 **structured skill bundles** — directory-based capability packs the agent
@@ -1024,6 +1042,7 @@ After installation, `codeep` is available globally in your terminal. Simply run
 |---------|-------------|
 | `/sessions` | List and load saved sessions |
 | `/sessions delete <name>` | Delete a specific session |
+| `/cloud` | List and resume sessions synced from other devices |
 | `/rename <name>` | Rename current session |
 | `/search <term>` | Search through chat history |
 | `/export` | Export chat to MD/JSON/TXT format |
@@ -1032,7 +1051,8 @@ After installation, `codeep` is available globally in your terminal. Simply run
 
 | Command | Description |
 |---------|-------------|
-| `/apply` | Apply file changes from AI response |
+| `/apply` | Apply file changes from AI response (shows hunk counts; `--only file.ts:0,1` for selective apply, `--interactive`/`-i` for per-hunk review) |
+| `/web-cache` | Show `@web` fetch cache stats (alias `/web-cache clear` to reset) |
 | `/copy [n]` | Copy code block to clipboard (n = block number, -1 = last) |
 | `/paste` | Paste content from clipboard into chat |
 | `/add <path>` | Add file(s) to conversation context |

package/dist/acp/commands.js

@@ -1023,8 +1023,29 @@ Anything else the agent should know — edge cases, gotchas, things to double-ch
         }
         case 'mcp': {
             const sub = args[0]?.toLowerCase();
-            const { addProjectMcpServer, removeProjectMcpServer, loadMcpServerConfig } = await import('../utils/mcpConfig.js');
+            const { addProjectMcpServer, removeProjectMcpServer, loadMcpServerConfig, loadMcpServerConfigSplit, isWorkspaceMcpTrusted, trustWorkspaceMcp, untrustWorkspaceMcp } = await import('../utils/mcpConfig.js');
             const { registerSessionServers } = await import('../utils/mcpRegistry.js');
+            if (sub === 'trust') {
+                if (isWorkspaceMcpTrusted(session.workspaceRoot)) {
+                    return { handled: true, response: '_Workspace MCP servers are already trusted here._' };
+                }
+                trustWorkspaceMcp(session.workspaceRoot);
+                const { workspace } = loadMcpServerConfigSplit(session.workspaceRoot);
+                if (workspace.length === 0) {
+                    return { handled: true, response: '_Workspace trusted — no workspace MCP servers defined yet._' };
+                }
+                onChunk(`_Workspace trusted. Spawning ${workspace.length} MCP server(s)…_\n\n`);
+                const merged = loadMcpServerConfig(session.workspaceRoot);
+                const { registered, errors } = await registerSessionServers(session.sessionId, merged, { workspaceRoot: session.workspaceRoot });
+                const lines = [`Trusted workspace MCP servers (${registered.length} tool(s) available).`];
+                for (const e of errors)
+                    lines.push(`- \`${e.server}\` failed: ${e.error}`);
+                return { handled: true, response: lines.join('\n') };
+            }
+            if (sub === 'untrust') {
+                untrustWorkspaceMcp(session.workspaceRoot);
+                return { handled: true, response: '_Workspace MCP trust revoked — workspace servers won\'t spawn for new sessions._' };
+            }
             if (sub === 'add') {
                 // /mcp add <name> <command> [args...]
                 const name = args[1];

package/dist/acp/server.js

@@ -8,7 +8,7 @@ import { StdioTransport } from './transport.js';
 import { runAgentSession } from './session.js';
 import { loadCustomCommands } from '../utils/customCommands.js';
 import { registerSessionServers, disposeAllSessions as disposeAllMcpSessions } from '../utils/mcpRegistry.js';
-import { loadMcpServerConfig, mergeMcpServers } from '../utils/mcpConfig.js';
+import { loadMcpServerConfigSplit, isWorkspaceMcpTrusted, mergeMcpServers } from '../utils/mcpConfig.js';
 import { handleMcpSamplingRequest } from '../utils/mcpSamplingBridge.js';
 import { executeCommandAsync } from '../utils/shell.js';
 import { initWorkspace, loadWorkspace, handleCommand } from './commands.js';
@@ -539,7 +539,18 @@ export function startAcpServer() {
      * in-flight registration so tool calls don't race the startup.
      */
     function spawnMcpServersForSession(acpSessionId, cwd, acpServers, label) {
-        const fromConfig = loadMcpServerConfig(cwd);
+        // Workspace-sourced servers (.codeep/mcp_servers.json, .mcp.json) travel
+        // with the repo, so they spawn only for workspaces the user has trusted
+        // (same gate the TUI prompts for at startup). ACP-provided servers are
+        // the editor's own config and global ~/.codeep entries are the user's —
+        // both spawn unconditionally.
+        const { global: globalServers, workspace: workspaceServers } = loadMcpServerConfigSplit(cwd);
+        const workspaceTrusted = isWorkspaceMcpTrusted(cwd);
+        if (workspaceServers.length > 0 && !workspaceTrusted) {
+            process.stderr.write(`[codeep-acp] MCP (${label}): skipped ${workspaceServers.length} workspace server(s) — untrusted workspace. ` +
+                `Run /mcp trust (or \`codeep\` in the repo once) to enable.\n`);
+        }
+        const fromConfig = workspaceTrusted ? [...globalServers, ...workspaceServers] : globalServers;
         const merged = mergeMcpServers(fromConfig, acpServers);
         if (merged.length === 0)
             return;

package/dist/acp/session.js

@@ -87,11 +87,32 @@ export function buildRawOutput(toolName, params, toolResult) {
 }
 export async function runAgentSession(opts) {
     const projectContext = buildProjectContext(opts.workspaceRoot);
+    // Expand `@folder`/`@file`/`@git` mentions in the user prompt, the
+    // same way the TUI does (see src/renderer/main.ts). ACP clients
+    // (VS Code / Zed) get the same inline-context UX. `@web` is async
+    // and fetched here too.
+    let enrichedPrompt = opts.prompt;
+    try {
+        const { expandFileAndFolderMentions, expandGitMentions } = await import('../utils/mentions.js');
+        const { expandWebMentions } = await import('../utils/webFetch.js');
+        const fileResult = expandFileAndFolderMentions(enrichedPrompt, { root: opts.workspaceRoot });
+        const gitResult = await expandGitMentions(fileResult.enrichedPrompt, { root: opts.workspaceRoot });
+        const webResult = await expandWebMentions(gitResult.enrichedPrompt);
+        enrichedPrompt = webResult.enrichedPrompt;
+        // Surface failures as thoughts so the editor shows them.
+        const allFailures = [...fileResult.failures, ...gitResult.failures, ...webResult.failures];
+        if (allFailures.length > 0 && opts.onThought) {
+            opts.onThought(allFailures.map((f) => `${f.mention}: ${f.reason}`).join(' · '));
+        }
+    }
+    catch {
+        // Mention expansion is best-effort — never block the agent run.
+    }
     let toolCallCounter = 0;
     // Maps tool call key → ACP toolCallId so onToolResult can emit finished/error status
     const toolCallIdMap = new Map();
     let chunksEmitted = 0;
-    const result = await runAgent(opts.prompt, projectContext, {
+    const result = await runAgent(enrichedPrompt, projectContext, {
         abortSignal: opts.abortSignal,
         onChunk: (text) => { chunksEmitted++; opts.onChunk(text); },
         onIteration: (_iteration, message) => {

package/dist/config/index.d.ts

@@ -140,6 +140,16 @@ export interface ConfigSchema {
 }
 export type { AgentMode };
 export type { LanguageCode };
+/**
+ * Get sessions directory - local .codeep/sessions/ if in project, otherwise global
+ */
+export declare function getSessionsDir(projectPath?: string): string;
+/**
+ * Check if directory is a project
+ * Looks for common project indicators: package.json, pyproject.toml, Cargo.toml, go.mod, composer.json, etc.
+ * Also checks if user has manually initialized this folder as a project (.codeep/project.json)
+ */
+export declare function isProjectDirectory(path: string): boolean;
 /**
  * Check if directory has standard project markers (not manually initialized)
  */

package/dist/config/index.js

@@ -9,7 +9,7 @@ import { createSecureStorage } from '../utils/keychain.js';
 /**
  * Get sessions directory - local .codeep/sessions/ if in project, otherwise global
  */
-function getSessionsDir(projectPath) {
+export function getSessionsDir(projectPath) {
     if (projectPath && isProjectDirectory(projectPath)) {
         const localDir = join(projectPath, '.codeep', 'sessions');
         if (!existsSync(localDir)) {
@@ -36,7 +36,7 @@ function getLocalConfigPath(projectPath) {
  * Looks for common project indicators: package.json, pyproject.toml, Cargo.toml, go.mod, composer.json, etc.
  * Also checks if user has manually initialized this folder as a project (.codeep/project.json)
  */
-function isProjectDirectory(path) {
+export function isProjectDirectory(path) {
     // Check if user has manually initialized this as a project
     const manualProjectMarker = join(path, '.codeep', 'project.json');
     if (existsSync(manualProjectMarker)) {

package/dist/config/providers.js

@@ -232,18 +232,21 @@ export const PROVIDERS = {
             openai: { baseUrl: 'https://api.moonshot.ai/v1', authHeader: 'Bearer', supportsNativeTools: true },
         },
         models: [
-            { id: 'kimi-k2.7-code', name: 'Kimi K2.7 Code', description: 'Flagship agentic coding model (256K context)' },
+            { id: 'kimi-k3-code', name: 'Kimi K3 Code', description: 'Newest flagship agentic coding model (1M context, deep reasoning)' },
+            { id: 'kimi-k3-code-highspeed', name: 'Kimi K3 Code (High-Speed)', description: 'Throughput-tuned K3 Code for latency-sensitive loops' },
+            { id: 'kimi-k3-thinking', name: 'Kimi K3 Thinking', description: 'K3 with explicit reasoning traces (highest quality, slower)' },
+            { id: 'kimi-k2.7-code', name: 'Kimi K2.7 Code', description: 'Previous-gen flagship agentic coding model (256K context)' },
             { id: 'kimi-k2.7-code-highspeed', name: 'Kimi K2.7 Code (High-Speed)', description: 'Throughput-tuned K2.7 Code for latency-sensitive loops' },
             { id: 'kimi-k2.6', name: 'Kimi K2.6', description: 'Previous-gen multimodal reasoning model' },
             { id: 'kimi-k2.5', name: 'Kimi K2.5', description: 'Older general-purpose model (cheaper)' },
         ],
-        defaultModel: 'kimi-k2.7-code',
+        defaultModel: 'kimi-k3-code',
         defaultProtocol: 'openai',
-        maxOutputTokens: 32_768,
+        maxOutputTokens: 65_536,
         envKey: 'MOONSHOT_API_KEY',
         subscribeUrl: 'https://platform.kimi.ai/console/api-keys',
         groupLabel: 'Kimi — API (pay-per-use)',
-        hint: 'Pay-per-use via Moonshot API key (platform.kimi.ai).',
+        hint: 'Pay-per-use via Moonshot API key (platform.kimi.ai). K3 models support 1M context and explicit reasoning.',
     },
     'kimi-cn': {
         name: 'Kimi China (Moonshot)',
@@ -252,18 +255,21 @@ export const PROVIDERS = {
             openai: { baseUrl: 'https://api.moonshot.cn/v1', authHeader: 'Bearer', supportsNativeTools: true },
         },
         models: [
-            { id: 'kimi-k2.7-code', name: 'Kimi K2.7 Code', description: 'Flagship agentic coding model (256K context)' },
+            { id: 'kimi-k3-code', name: 'Kimi K3 Code', description: 'Newest flagship agentic coding model (1M context, deep reasoning)' },
+            { id: 'kimi-k3-code-highspeed', name: 'Kimi K3 Code (High-Speed)', description: 'Throughput-tuned K3 Code' },
+            { id: 'kimi-k3-thinking', name: 'Kimi K3 Thinking', description: 'K3 with explicit reasoning traces' },
+            { id: 'kimi-k2.7-code', name: 'Kimi K2.7 Code', description: 'Previous-gen flagship agentic coding model (256K context)' },
             { id: 'kimi-k2.7-code-highspeed', name: 'Kimi K2.7 Code (High-Speed)', description: 'Throughput-tuned K2.7 Code' },
             { id: 'kimi-k2.6', name: 'Kimi K2.6', description: 'Previous-gen multimodal reasoning model' },
             { id: 'kimi-k2.5', name: 'Kimi K2.5', description: 'Older general-purpose model' },
         ],
-        defaultModel: 'kimi-k2.7-code',
+        defaultModel: 'kimi-k3-code',
         defaultProtocol: 'openai',
-        maxOutputTokens: 32_768,
+        maxOutputTokens: 65_536,
         envKey: 'MOONSHOT_CN_API_KEY',
         subscribeUrl: 'https://platform.moonshot.cn/console/api-keys',
         groupLabel: 'Kimi China — API (pay-per-use)',
-        hint: 'Pay-per-use via Moonshot China API key (platform.moonshot.cn).',
+        hint: 'Pay-per-use via Moonshot China API key (platform.moonshot.cn). K3 models support 1M context.',
     },
     // ── Grok (xAI) ────────────────────────────────────────────────────
     // Pay-per-use today (console.x.ai key). The SuperGrok / X Premium+
@@ -276,8 +282,9 @@ export const PROVIDERS = {
             openai: { baseUrl: 'https://api.x.ai/v1', authHeader: 'Bearer', supportsNativeTools: true },
         },
         models: [
+            { id: 'grok-4.5', name: 'Grok 4.5', description: 'Flagship reasoning model — highest quality, 500K context' },
             { id: 'grok-build-0.1', name: 'Grok Build 0.1', description: 'Agentic coding model — fast, 256K context' },
-            { id: 'grok-4.3', name: 'Grok 4.3', description: 'Flagship — highest quality, 1M context' },
+            { id: 'grok-4.3', name: 'Grok 4.3', description: 'Previous flagship, 1M context' },
             { id: 'grok-code-fast-1', name: 'Grok Code Fast 1', description: 'Low-cost speed-first coder (alias of Build 0.1)' },
             { id: 'grok-4-fast-reasoning', name: 'Grok 4 Fast (reasoning)', description: 'Cheap reasoning model, very large context' },
         ],
@@ -302,7 +309,7 @@ export const PROVIDERS = {
         models: [
             { id: 'qwen3-coder-plus', name: 'Qwen3-Coder Plus', description: 'Flagship coding model — best quality' },
             { id: 'qwen3-coder-next', name: 'Qwen3-Coder Next', description: 'Balanced quality/speed/cost' },
-            { id: 'qwen3-max', name: 'Qwen3-Max', description: 'Flagship general model (code + reasoning)' },
+            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Flagship general model (code + reasoning)' },
         ],
         defaultModel: 'qwen3-coder-plus',
         defaultProtocol: 'openai',
@@ -323,7 +330,7 @@ export const PROVIDERS = {
             { id: 'qwen3-coder-plus', name: 'Qwen3-Coder Plus', description: 'Flagship coding model (256K, up to 1M)' },
             { id: 'qwen3-coder-next', name: 'Qwen3-Coder Next', description: 'Balanced quality/speed/cost' },
             { id: 'qwen3-coder-flash', name: 'Qwen3-Coder Flash', description: 'Fast/cheap coder' },
-            { id: 'qwen3-max', name: 'Qwen3-Max', description: 'Flagship general model' },
+            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Flagship general model' },
         ],
         defaultModel: 'qwen3-coder-plus',
         defaultProtocol: 'openai',
@@ -343,7 +350,7 @@ export const PROVIDERS = {
         models: [
             { id: 'qwen3-coder-plus', name: 'Qwen3-Coder Plus', description: 'Flagship coding model — best quality' },
             { id: 'qwen3-coder-next', name: 'Qwen3-Coder Next', description: 'Balanced quality/speed/cost' },
-            { id: 'qwen3-max', name: 'Qwen3-Max', description: 'Flagship general model' },
+            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Flagship general model' },
         ],
         defaultModel: 'qwen3-coder-plus',
         defaultProtocol: 'openai',
@@ -364,7 +371,7 @@ export const PROVIDERS = {
             { id: 'qwen3-coder-plus', name: 'Qwen3-Coder Plus', description: 'Flagship coding model' },
             { id: 'qwen3-coder-next', name: 'Qwen3-Coder Next', description: 'Balanced quality/speed/cost' },
             { id: 'qwen3-coder-flash', name: 'Qwen3-Coder Flash', description: 'Fast/cheap coder' },
-            { id: 'qwen3-max', name: 'Qwen3-Max', description: 'Flagship general model' },
+            { id: 'qwen3.7-max', name: 'Qwen3.7-Max', description: 'Flagship general model' },
         ],
         defaultModel: 'qwen3-coder-plus',
         defaultProtocol: 'openai',
@@ -404,11 +411,14 @@ export const PROVIDERS = {
             },
         },
         models: [
-            { id: 'gpt-5.5', name: 'GPT-5.5', description: 'Latest GPT model — best for coding' },
-            { id: 'gpt-5.4', name: 'GPT-5.4', description: 'Previous generation GPT' },
+            { id: 'gpt-5.6-sol', name: 'GPT-5.6 Sol', description: 'Most capable GPT — best for coding & agentic work' },
+            { id: 'gpt-5.6-terra', name: 'GPT-5.6 Terra', description: 'Balanced — GPT-5.5 quality at about half the price' },
+            { id: 'gpt-5.6-luna', name: 'GPT-5.6 Luna', description: 'Fast and cheap — high-volume workloads' },
+            { id: 'gpt-5.5', name: 'GPT-5.5', description: 'Previous flagship GPT' },
+            { id: 'gpt-5.4', name: 'GPT-5.4', description: 'Older generation GPT' },
             { id: 'gpt-5.4-mini', name: 'GPT-5.4 Mini', description: 'Faster and cheaper GPT-5.4' },
         ],
-        defaultModel: 'gpt-5.5',
+        defaultModel: 'gpt-5.6-sol',
         defaultProtocol: 'openai',
         useMaxCompletionTokens: true,
         requiresDefaultTemperature: true,
@@ -429,11 +439,11 @@ export const PROVIDERS = {
         },
         models: [
             { id: 'claude-fable-5', name: 'Claude Fable 5', description: 'Most capable — hardest reasoning & long-horizon agentic work' },
-            { id: 'claude-opus-4-8', name: 'Claude Opus 4.8', description: 'Most capable Opus model' },
+            { id: 'claude-opus-5', name: 'Claude Opus 5', description: 'Complex agentic coding & deep reasoning — the Opus workhorse' },
             { id: 'claude-sonnet-5', name: 'Claude Sonnet 5', description: 'Best balance of speed and intelligence' },
             { id: 'claude-haiku-4-5-20251001', name: 'Claude Haiku', description: 'Fastest and most affordable' },
         ],
-        defaultModel: 'claude-opus-4-8',
+        defaultModel: 'claude-opus-5',
         defaultProtocol: 'anthropic',
         envKey: 'ANTHROPIC_API_KEY',
         groupLabel: 'Anthropic',
@@ -452,6 +462,7 @@ export const PROVIDERS = {
         models: [
             { id: 'gemini-3.1-pro-preview', name: 'Gemini 3.1 Pro', description: 'Most capable Gemini model' },
             { id: 'gemini-3.5-flash', name: 'Gemini 3.5 Flash', description: 'Pro-level intelligence at Flash speed' },
+            { id: 'gemini-3.1-flash-lite', name: 'Gemini 3.1 Flash-Lite', description: 'Low-latency, low-cost workhorse (1M context)' },
         ],
         defaultModel: 'gemini-3.1-pro-preview',
         defaultProtocol: 'openai',
@@ -479,7 +490,7 @@ export const PROVIDERS = {
             { id: 'anthropic/claude-fable-5', name: 'Claude Fable 5', description: 'Anthropic — most capable' },
             { id: 'anthropic/claude-opus-4', name: 'Claude Opus 4', description: 'Anthropic — Opus tier' },
             { id: 'anthropic/claude-sonnet-5', name: 'Claude Sonnet 5', description: 'Anthropic — balanced' },
-            { id: 'openai/gpt-5.5', name: 'GPT-5.5', description: 'OpenAI — flagship' },
+            { id: 'openai/gpt-5.6-sol', name: 'GPT-5.6 Sol', description: 'OpenAI — flagship' },
             { id: 'openai/gpt-5.4-mini', name: 'GPT-5.4 Mini', description: 'OpenAI — fast/cheap' },
             { id: 'google/gemini-3.1-pro', name: 'Gemini 3.1 Pro', description: 'Google — multimodal' },
             { id: 'meta-llama/llama-3.1-405b-instruct', name: 'Llama 3.1 405B', description: 'Meta — open weights, largest' },
@@ -487,7 +498,7 @@ export const PROVIDERS = {
             { id: 'deepseek/deepseek-v4', name: 'DeepSeek V4', description: 'DeepSeek via OpenRouter' },
             { id: 'mistralai/mistral-large', name: 'Mistral Large', description: 'Mistral — flagship' },
             { id: 'qwen/qwen-2.5-coder-32b-instruct', name: 'Qwen 2.5 Coder 32B', description: 'Alibaba — coding-tuned' },
-            { id: 'x-ai/grok-2', name: 'Grok 2', description: 'xAI via OpenRouter' },
+            { id: 'x-ai/grok-4.5', name: 'Grok 4.5', description: 'xAI — flagship reasoning' },
         ],
         defaultModel: 'anthropic/claude-opus-4',
         defaultProtocol: 'openai',
@@ -661,8 +672,8 @@ export function providerNoStreamWithTools(providerId) {
  * internally and 400 on any custom value, so they're here too.
  */
 const SAMPLING_PARAMS_REJECTED = [
-    'claude-fable-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5',
-    'kimi-k2.7-code', 'kimi-for-coding',
+    'claude-fable-5', 'claude-opus-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5',
+    'kimi-k3-code', 'kimi-k3-thinking', 'kimi-k2.7-code', 'kimi-for-coding',
 ];
 export function modelRejectsSamplingParams(model) {
     return SAMPLING_PARAMS_REJECTED.some(id => model === id || model.startsWith(`${id}-`));
@@ -704,10 +715,10 @@ export function modelSupportsReasoningEffort(providerId, model) {
     const id = canonicalModelId(model);
     switch (providerId) {
         case 'anthropic':
-            // Effort is GA on Opus 4.5+, Sonnet 4.6/5, Fable 5 — NOT Haiku or Sonnet 4.5.
+            // Effort is GA on Opus 5, Opus 4.5+, Sonnet 4.6/5, Fable 5 — NOT Haiku or Sonnet 4.5.
             if (idMatches(id, 'claude-haiku-4-5') || idMatches(id, 'claude-sonnet-4-5'))
                 return false;
-            return /^claude-(opus-4-([5-9]|\d\d)|sonnet-(4-6|5)|fable-5)/.test(id);
+            return /^claude-(opus-5|opus-4-([5-9]|\d\d)|sonnet-(4-6|5)|fable-5)/.test(id);
         case 'openai':
             // GPT-5.x are reasoning models — reasoning_effort across the family (incl. mini).
             return id.startsWith('gpt-5');

package/dist/renderer/App.d.ts

@@ -20,6 +20,37 @@ export interface ConfirmOptions {
     onConfirm: () => void;
     onCancel?: () => void;
 }
+/**
+ * One hunk in the interactive `/apply --interactive` picker.
+ * `lines` are already-formatted diff lines (e.g. `+ added`, `- removed`).
+ */
+export interface HunkPickerItem {
+    /** File path this hunk belongs to. */
+    path: string;
+    /** 0-based hunk index within the file diff. */
+    hunkIndex: number;
+    /** Human-readable hunk header, e.g. `@@ -12,3 +12,5 @@`. */
+    header: string;
+    /** Pre-formatted diff lines to display. */
+    lines: string[];
+}
+/**
+ * Options for the interactive hunk picker. The picker walks the user
+ * through `items` one at a time; for each they accept (`y`/Enter) or
+ * skip (`n`). `a` accepts all remaining, `q`/Esc quits.
+ *
+ * `onComplete` fires once with the set of accepted `[path, hunkIndex]`
+ * pairs (possibly empty) so the caller can apply them via
+ * `applyHunksToFiles`.
+ */
+export interface HunkPickerOptions {
+    title: string;
+    items: HunkPickerItem[];
+    onComplete: (accepted: Array<{
+        path: string;
+        hunkIndex: number;
+    }>) => void;
+}
 export interface AppOptions {
     onSubmit: (message: string) => Promise<void>;
     onCommand: (command: string, args: string[]) => void;
@@ -29,6 +60,8 @@ export interface AppOptions {
     getStatus: () => StatusInfo;
     hasWriteAccess?: () => boolean;
     hasProjectContext?: () => boolean;
+    /** Project root for `@mention` autocomplete suggestions. Falls back to cwd. */
+    getProjectRoot?: () => string;
 }
 export declare class App {
     private screen;
@@ -67,9 +100,19 @@ export declare class App {
     private showAutocomplete;
     private autocompleteIndex;
     private autocompleteItems;
+    private showMentionAutocomplete;
+    private mentionIndex;
+    private mentionItems;
+    private mentionAtStart;
+    /** Project root for resolving `suggestMentions`. Cached per update. */
+    private mentionRoot;
     private confirmOpen;
     private confirmOptions;
     private confirmSelection;
+    private hunkPickerOpen;
+    private hunkPickerOptions;
+    private hunkPickerIndex;
+    private hunkPickerAccepted;
     private menuOpen;
     private menuTitle;
     /** Filtered view shown to the user; derived from `menuItemsAll` + `menuFilter`. */
@@ -231,6 +274,11 @@ export declare class App {
      * Show confirmation dialog
      */
     showConfirm(options: ConfirmOptions): void;
+    /**
+     * Show the interactive hunk picker (`/apply --interactive`).
+     * The caller passes pre-built items + an `onComplete` callback.
+     */
+    showHunkPicker(options: HunkPickerOptions): void;
     /**
      * Show permission dialog (inline, below status bar)
      */
@@ -314,6 +362,13 @@ export declare class App {
      * Update autocomplete suggestions
      */
     private updateAutocomplete;
+    /**
+     * Replace the in-progress `@query` (from `mentionAtStart` to the
+     * cursor) with the selected mention's path. Keeps the `@` prefix and
+     * positions the cursor right after the inserted path so the user can
+     * keep typing the rest of the message.
+     */
+    private applyMentionSelection;
     /**
      * Handle inline status keys
      */
@@ -352,6 +407,15 @@ export declare class App {
     private handleInlinePermissionKey;
     private handleInlineSessionPickerKey;
     private handleInlineConfirmKey;
+    /**
+     * Handle keys in the interactive hunk picker.
+     *   y / Enter / →  accept this hunk, advance
+     *   n / ←         skip this hunk, advance
+     *   a             accept this + all remaining, finish
+     *   q / Esc       finish without accepting this hunk
+     *   ↑ / ↓         navigate (preview only — no decision)
+     */
+    private handleHunkPickerKey;
     /**
      * Submit the current input buffer (used by Enter and Escape-in-multiline)
      */
@@ -373,6 +437,11 @@ export declare class App {
      * Render inline confirmation dialog below status bar
      */
     private renderInlineConfirm;
+    /**
+     * Render inline hunk picker (`/apply --interactive`).
+     * Shows the current hunk's diff + the y/n/a/q key legend.
+     */
+    private renderInlineHunkPicker;
     /**
      * Render input line
      */
@@ -394,6 +463,14 @@ export declare class App {
      * Render inline autocomplete below status bar
      */
     private renderInlineAutocomplete;
+    /**
+     * Render inline `@mention` file picker below the status bar.
+     *
+     * Mirrors the layout of `renderInlineAutocomplete` (separator → title →
+     * items → footer) but shows file paths with their parent directory as
+     * the description, and a `@` prefix instead of `/`.
+     */
+    private renderInlineMentionPicker;
     /**
      * Render inline permission dialog
      */
@@ -413,15 +490,6 @@ export declare class App {
     /**
      * Get color for action type
      */
-    private getActionColor;
-    /**
-     * Format action target for display
-     */
-    private formatActionTarget;
-    /**
-     * Get action label for display
-     */
-    private getActionLabel;
     /**
      * Render status bar
      */
@@ -430,23 +498,6 @@ export declare class App {
      * Get visible messages (including streaming)
      */
     private getVisibleMessages;
-    /**
-     * Format message into lines with syntax highlighting for code blocks
-     */
-    private formatWelcomeMessage;
-    private formatMessage;
-    /**
-     * Apply inline markdown formatting (bold, italic, inline code) to a line
-     */
-    private applyInlineMarkdown;
-    /**
-     * Format plain text lines with markdown support
-     */
-    private formatTextLines;
-    /**
-     * Format code block with syntax highlighting (no border)
-     */
-    private formatCodeBlock;
     /**
      * Render inline search screen
      */
@@ -471,8 +522,4 @@ export declare class App {
      * Get decrypted logo for intro animation
      */
     private getDecryptedLogo;
-    /**
-     * Word wrap
-     */
-    private wordWrap;
 }