v2.9.64

.github/workflows/release.yml

@@ -61,12 +61,27 @@ jobs:
         with:
           path: dist-vsix
           merge-multiple: true
+      - name: Resolve release tag from package.json
+        id: meta
+        run: |
+          version="$(node -p "require('./package.json').version")"
+          if [ -z "$version" ]; then
+            echo "::error::Could not read version from package.json"
+            exit 1
+          fi
+          echo "version=$version" >> "$GITHUB_OUTPUT"
+          echo "tag=v${version}" >> "$GITHUB_OUTPUT"
       - name: Generate release notes
         run: cp CHANGELOG.md release-notes.md
-      - uses: softprops/action-gh-release@v2
+      # workflow_dispatch has no tag ref; create/update the GitHub Release from package.json.
+      - uses: softprops/action-gh-release@v3
         with:
+          tag_name: ${{ steps.meta.outputs.tag }}
+          name: ${{ steps.meta.outputs.tag }}
+          target_commitish: ${{ github.sha }}
           body_path: release-notes.md
           files: 'dist-vsix/*.vsix'
+          fail_on_unmatched_files: true
 
   publish-marketplace:
     needs: package

.vscode/settings.json

@@ -27,5 +27,7 @@
     "wasms",
     "xenova"
   ],
-  "mitii.onboarding.completed": true
+  "mitii.onboarding.completed": true,
+  "mitii.provider.model": "my-qwen-64k:latest",
+  "mitii.provider.contextWindow": 45000
 }

README.md

@@ -15,7 +15,7 @@
   <a href="LICENSE"><img alt="License: AGPL v3" src="https://img.shields.io/badge/License-AGPL_v3-blue.svg"></a>
   <a href="https://code.visualstudio.com/"><img alt="VS Code 1.124+" src="https://img.shields.io/badge/VS%20Code-1.124%2B-007ACC?logo=visualstudiocode"></a>
   <a href="https://nodejs.org/"><img alt="Node 20+" src="https://img.shields.io/badge/Node-20%2B-339933?logo=node.js"></a>
-  <img alt="Version 2.9.57" src="https://img.shields.io/badge/version-2.9.57-111111">
+  <img alt="Version 2.9.66" src="https://img.shields.io/badge/version-2.9.66-111111">
   <a href="https://docs.mitii.dev"><img alt="Documentation" src="https://img.shields.io/badge/docs-docs.mitii.dev-ff751f"></a>
 </p>
 

apps/acp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mitii/acp",
-  "version": "2.9.57",
+  "version": "2.9.66",
   "description": "Mitii ACP-lite stdio bridge (Phase 3). Decision Policy remains authority; V8 does not import ACP.",
   "license": "AGPL-3.0-or-later",
   "type": "module",

apps/cli/README.md

@@ -71,7 +71,7 @@ mitii export-session "Summarize this repo" --out session.json --echo
 | `run --auto "<task>"` | Unattended CI run (agent + apply autonomy; no prompts) |
 | `session` | Interactive prompt loop with MITII banner |
 | `index` | Full workspace index + publish repository state |
-| `review` | Deterministic review prep / SARIF (`--preview`, `--from`/`--to`, `--commit`, `--format`, `--output`). For LLM findings use `mitii ask … --skill code-review-and-quality` or VS Code **Review** mode |
+| `review` | Deterministic review prep / SARIF (`--preview`, `--from`/`--to`, `--commit`, `--format`, `--output`). For LLM findings use `mitii ask … --skill code-review-and-quality` (recipe/skill) or the VS Code **Review** button for working-tree changes |
 | `status` | Show latest persisted repository state |
 | `export-session` | Run ask and write secret-free JSON export |
 | `connect` | Bridge Mitii into Telegram, Discord, or Slack |

apps/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mitii/cli",
-  "version": "2.9.57",
+  "version": "2.9.66",
   "description": "Mitii headless CLI over @mitii/sdk. Phase 0: --origin/--autonomy/--agent for CI automation.",
   "license": "AGPL-3.0-or-later",
   "publishConfig": {

apps/cli/src/help.ts

@@ -46,7 +46,7 @@ Commands:
                    --format json|sarif Output shape (default json)
                    --output <path>     Write result to file
                    --effort low|medium|high  Prep effort band
-                   Full LLM review: VS Code Review mode, or:
+                   Full LLM review: VS Code Review button (git changes), or:
                      mitii ask "review these changes" --mode ask --skill code-review-and-quality
   status           Show latest persisted repository state
   export-session   Run ask and write secret-free JSON export

apps/daemon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mitii/daemon",
-  "version": "2.9.57",
+  "version": "2.9.66",
   "description": "Mitii automation daemon process entry (Phase 1). Long-lived schedule runner.",
   "license": "AGPL-3.0-or-later",
   "type": "module",

apps/vscode/README.md

@@ -33,7 +33,7 @@ For cloud providers, run **Mitii: Set Provider API Key** (stored in VS Code Secr
 
 - **Repository-aware context** - SQLite FTS5, symbols, optional vectors, repo map, diagnostics, Git state, and `@` attachments
 - **Skills** - force-attach playbooks with `/` or `@skill:id` in chat (up to 3 per message); workspace skills in `.mitii/skills/`
-- **Ask / Plan / Agent** - read-only Q&A, structured plans, and controlled edits. Use the collapsible **Review** bar above chat (file count → Review) for structured working-tree findings via `emit_review_finding` — not a fourth mode. After findings appear, use **Dismiss**, **Fix**, or **Fix all** (Agent + `fix-review-findings` recipe).
+- **Ask / Plan / Agent** - read-only Q&A, structured plans, and controlled edits. The composer **Review** strip lists **this chat’s** Mitii file changes (expand with **Review**). Enable **Settings → Features → Code Review** to add **Code Review (N)** over the full git working tree, then severity-tagged findings with **Fix** / **Fix all**. Free-form “code review” chat is not a structured review run.
 - **FIM autocomplete** - optional inline ghost text from a low-latency OpenAI-compatible `prompt` + `suffix` endpoint
 - **Safety** - configurable approvals, path containment, command policy, pre-write checkpoints, workspace trust
 - **Providers** - Echo, Anthropic (Claude), Gemini, and OpenAI-compatible endpoints (DeepSeek, OpenRouter, Azure, Ollama, custom `/v1`)
@@ -45,7 +45,7 @@ For cloud providers, run **Mitii: Set Provider API Key** (stored in VS Code Secr
 | Command | Purpose |
 |---|---|
 | **Mitii: Open Chat** | Open the sidebar |
-| **Mitii: Review Working Tree Changes** | Switch to Review mode and run a structured review |
+| **Mitii: Review Working Tree Changes** | Expand the chat review strip. **Review** = this-chat Mitii edits. With Settings → Features → Code Review on, **Code Review (N)** runs LLM analysis over all `N` git changes |
 | **Mitii: Index Workspace** | Rebuild repository index |
 | **Mitii: Show Settings** | Provider, index, MCP, workspace |
 | **Mitii: Toggle Autocomplete** | Enable or disable FIM inline suggestions |

apps/vscode/package.json

@@ -2,7 +2,7 @@
   "name": "mitii-ai-agent",
   "displayName": "Mitii AI Agent",
   "description": "Local-first VS Code AI coding agent with repository-aware context and controlled execution",
-  "version": "2.9.57",
+  "version": "2.9.66",
   "publisher": "mitii",
   "license": "AGPL-3.0-or-later",
   "icon": "media/mitii-logo.png",
@@ -1028,6 +1028,11 @@
           "default": true,
           "description": "Show live reasoning / thinking activity in the Mitii chat UI"
         },
+        "mitii.ui.features.codeReviewButton": {
+          "type": "boolean",
+          "default": false,
+          "description": "Show the Code Review button on the working-tree bar. Review always lists git changes; Code Review runs an LLM analysis. Enable under Settings → Features."
+        },
         "mitii.ui.reasoningPreviewMaxChars": {
           "type": "number",
           "default": 8000,
@@ -1365,6 +1370,33 @@
               }
             }
           }
+        },
+        "mitii.tokenBudget.visiblePlanMinUsableRatio": {
+          "type": "number",
+          "default": 0.35,
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Visible-plan affordability threshold as a fraction of the context window."
+        },
+        "mitii.tokenBudget.changeImpactMinUsableRatio": {
+          "type": "number",
+          "default": 0.35,
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Change-impact affordability threshold as a fraction of the context window."
+        },
+        "mitii.tokenBudget.maxPatchesPerCallCap": {
+          "type": "number",
+          "default": 96,
+          "minimum": 1,
+          "description": "Maximum patches allowed in one mutation call."
+        },
+        "mitii.loopPolicy.maxStructuredReviewRecoveries": {
+          "type": "number",
+          "default": 2,
+          "minimum": 0,
+          "maximum": 8,
+          "description": "Recovery attempts when a structured review ends without findings (0 disables)."
         }
       }
     }

apps/vscode/src/automationHost.ts

@@ -1,5 +1,6 @@
 import { AutomationService } from '@mitii/automation';
 
+import { openSqliteDatabase } from './nativeSqlite.js';
 import type {
   AutomationRunView,
   AutomationSpecView,
@@ -56,7 +57,11 @@ export function handleAutomationHostMessage(input: {
   post: (msg: HostToWebviewMessage) => void;
   workspaceRoot: string | undefined;
 }): void {
-  const service = new AutomationService({});
+  const service = new AutomationService({
+    // Bundled better-sqlite3 has no default .node next to extension.js; use the
+    // Electron-staged binding under dist/native (same path as workspace index).
+    openDatabase: openSqliteDatabase,
+  });
   try {
     if (input.message.type === 'requestAutomations') {
       // Reconcile workspace file specs so the panel mirrors `mitii serve`.

apps/vscode/src/extension.ts

@@ -9,6 +9,7 @@ import {
   unwrapRecipeAnswer,
   type MitiiWritingRecipeId,
   IndexLockedError,
+  resolveMaximumIndexFiles,
 } from '@mitii/host';
 import { isSecurityConcern, WorkspaceIgnorePolicy } from '@mitii/v8';
 
@@ -378,6 +379,11 @@ export function activate(context: ExtensionContext): void {
           const snapshot = await buildWorkspaceSnapshot({
             workspaceRoot: root,
             workspaceId,
+            maxFiles: resolveMaximumIndexFiles(
+              vscode.workspace
+                .getConfiguration('mitii')
+                .get<number>('workspace.maximumIndexFiles'),
+            ),
           });
           fileCount = snapshot.fileCount;
           truncated = snapshot.truncated;
@@ -424,7 +430,8 @@ export function activate(context: ExtensionContext): void {
 
   const reviewChanges = async (): Promise<void> => {
     await vscode.commands.executeCommand('mitii.sidebar.focus');
-    sidebar?.post({ type: 'startReview', autoRun: true });
+    // Show git changes only — LLM Code Review is the separate bar button.
+    sidebar?.post({ type: 'startReview', autoRun: false });
   };
 
   const runWritingRecipe = async (

apps/vscode/src/hostAsk.ts

@@ -10,7 +10,7 @@ import {
   type RunEvent,
   type TaskList,
 } from '@mitii/sdk';
-import { loadUserSafetyRules } from '@mitii/host';
+import { loadUserSafetyRules, resolveMaximumIndexFiles } from '@mitii/host';
 import type * as vscode from 'vscode';
 
 import { formatDiagnosticsPromptBlock } from './context/diagnosticsContext.js';
@@ -166,6 +166,8 @@ export async function runAskInOutputChannel(options: {
   approvalMode?: string;
   pinnedPaths?: string[];
   requiredSkillIds?: string[];
+  /** Skills that must not be auto-matched for this run. */
+  excludedSkillIds?: string[];
   /** Attached MCP server ids for this turn (@mcp: / host pin). */
   requiredMcpServerIds?: string[];
   workspaceId?: string;
@@ -475,6 +477,9 @@ export async function runAskInOutputChannel(options: {
       ...(options.requiredSkillIds && options.requiredSkillIds.length > 0
         ? { requiredSkillIds: [...options.requiredSkillIds] }
         : {}),
+      ...(options.excludedSkillIds && options.excludedSkillIds.length > 0
+        ? { excludedSkillIds: [...options.excludedSkillIds] }
+        : {}),
       ...(options.requiredMcpServerIds &&
       options.requiredMcpServerIds.length > 0
         ? { requiredMcpServerIds: [...options.requiredMcpServerIds] }
@@ -739,6 +744,11 @@ async function autoPublishFullOrSnapshot(options: {
     const snap = await buildWorkspaceSnapshot({
       workspaceRoot: options.workspaceRoot,
       workspaceId: options.workspaceId,
+      maxFiles: resolveMaximumIndexFiles(
+        options.vs.workspace
+          .getConfiguration('mitii')
+          .get<number>('workspace.maximumIndexFiles'),
+      ),
     });
     await options.client.publishRepositoryState(snap.candidate);
     options.channel.appendLine(

apps/vscode/src/liveSandboxProcess.ts

@@ -0,0 +1,85 @@
+import type * as vscode from 'vscode';
+import type { ProcessPort, ProcessExecRequest } from '@mitii/v8';
+import {
+  createSandboxedProcessPort,
+  resolveSandboxPolicy,
+  resolveSandboxSettingsFromPreset,
+  type SandboxBackend,
+  type SandboxNetworkMode,
+} from '@mitii/host';
+
+type ResolvedSandboxSettings = {
+  enabled: boolean;
+  network: SandboxNetworkMode;
+};
+
+function readSandboxSettings(vs: typeof vscode): ResolvedSandboxSettings {
+  const cfg = vs.workspace.getConfiguration('mitii');
+  const configured = (key: string): boolean => {
+    const value = cfg.inspect(key);
+    return (
+      value?.globalValue !== undefined ||
+      value?.workspaceValue !== undefined ||
+      value?.workspaceFolderValue !== undefined
+    );
+  };
+  return resolveSandboxSettingsFromPreset({
+    approvalMode: cfg.get<string>('safety.approvalMode') ?? 'guided',
+    ...(configured('safety.sandbox.enabled')
+      ? { enabled: cfg.get<boolean>('safety.sandbox.enabled') === true }
+      : {}),
+    ...(configured('safety.sandbox.network')
+      ? { network: cfg.get<string>('safety.sandbox.network') ?? 'deny' }
+      : {}),
+  });
+}
+
+function settingsKey(settings: ResolvedSandboxSettings): string {
+  return `${settings.enabled ? 1 : 0}:${settings.network}`;
+}
+
+/**
+ * Process port that refreshes access policy when settings change, without
+ * rebuilding a wrapper on every exec.
+ *
+ * Once sandbox has been enabled for this port, it cannot be disabled through
+ * live settings (fail closed). Network may still tighten/loosen with approval
+ * mode so Guided ↔ Full access keeps working mid-session.
+ */
+export function createLiveSandboxedProcessPort(
+  vs: typeof vscode,
+  inner: ProcessPort,
+  workspaceRoot: string,
+  backend: SandboxBackend,
+): ProcessPort {
+  let cachedKey: string | undefined;
+  let cachedPort: ProcessPort | undefined;
+  let sandboxLatchedOn = false;
+
+  const resolvePort = (): ProcessPort => {
+    const raw = readSandboxSettings(vs);
+    const settings: ResolvedSandboxSettings = {
+      // Latch: after sandbox has run enabled, ignore live disable.
+      enabled: raw.enabled || sandboxLatchedOn,
+      network: raw.network,
+    };
+    if (settings.enabled) sandboxLatchedOn = true;
+
+    const key = settingsKey(settings);
+    if (cachedPort && cachedKey === key) return cachedPort;
+
+    cachedKey = key;
+    cachedPort = createSandboxedProcessPort(
+      inner,
+      resolveSandboxPolicy({ ...settings, workspaceRoot }),
+      backend,
+    );
+    return cachedPort;
+  };
+
+  return {
+    execFile(request: ProcessExecRequest) {
+      return resolvePort().execFile(request);
+    },
+  };
+}