mixdog 0.9.157 → 0.9.159

package/README.md

@@ -45,31 +45,35 @@ workflow setup.
 
 Terminal-Bench 2.1 — same model, same 89 tasks, same official verifier, with
 only the harness changed. Against the native CLI of each model family, Mixdog
-scores higher while spending less to get there.
+delivers the same results at the same speed — on a fraction of the context,
+for far less cost.
 
 ### GPT-5.6 Sol xhigh — Mixdog vs Codex CLI
 
 ![Terminal-Bench 2.1: Mixdog with GPT-5.6 Sol xhigh versus Codex CLI](https://raw.githubusercontent.com/tribgames/mixdog/main/benchmarks/terminal-bench-2.1/tb21-sol-vs-codex.svg)
 
-- **86.5%** (385/445) vs Codex CLI's **84.3%** (75/89)
-- **42%** lower priced cost — $0.641 vs $1.096 per trial
-- **45%** smaller median final context — 18.5k vs 33.5k tokens
-- **1.11×** faster — 339s vs 378s per trial
+- **39%** lower priced cost — $0.476 vs $0.782 per trial
+- **46%** smaller median final context — 18.5k vs 34.3k tokens
+- **86.5%** (385/445) vs Codex CLI's **86.1%** (383/445) — full `k=5` on both
+  sides, pass@5 **96.6%** vs 95.5%
+- Matched speed — 415s vs 437s wall time per trial
 
 ### Claude Opus 5 — Mixdog vs Claude Code
 
 ![Terminal-Bench 2.1: Mixdog with Claude Opus 5 versus Claude Code](https://raw.githubusercontent.com/tribgames/mixdog/main/benchmarks/terminal-bench-2.1/tb21-opus-vs-claude-code.svg)
 
-- **79/89** vs Claude Code's **77/89**
 - **19%** lower priced cost — $104.29 vs $129.21 per run
 - **28%** smaller median final context — 27.6k vs 38.2k tokens
-- **1.15×** faster
+- **79/89** vs Claude Code's **77/89**
+- **1.16×** faster — 610s vs 708s wall time per trial
 
-Every run uses the official Harbor verifier, fast mode off, a 272k context
-window, and zero retries. The Mixdog Sol run follows the protocol the official
-Terminal-Bench leaderboard requires — all 89 tasks repeated five times (`k=5`,
-445 trials); the Codex CLI baseline and both Opus-side runs are single passes
-(`k=1`, 89 trials each).
+Every run uses the official Harbor verifier, fast mode off, and a 272k context
+window; task failures and agent timeouts are never retried. The Sol comparison
+follows the protocol the official Terminal-Bench leaderboard requires on both
+sides — all 89 tasks repeated five times (`k=5`, 445 trials each); the
+Opus-side runs are single passes (`k=1`, 89 trials each). Speed is the full
+trial wall clock, and cost values both sides at the same current API list
+rates.
 
 The leaderboard is not accepting community submissions, so every run here ships
 its raw artifacts instead — Harbor verdicts, official verifier output, pinned
@@ -88,6 +92,13 @@ the harness, presets, and metric scripts that recompute each number above:
   MCP, skills, hooks, and plugins.
 - **Local memory** — semantic and lexical recall with project-scoped context
   and multilingual retrieval.
+- **Browser Use** — a logged-in Chromium pane the agent can drive: tabs,
+  forms, downloads, and snapshots, with one-time Chrome profile import
+  including cookies and passwords.
+- **Computer Use** — agent control of the Windows desktop through screen
+  capture, accessibility, OCR, and a strict guarded input contract.
+- **Office documents** — author and edit Word, Excel, and PowerPoint files
+  with model-authored design plans, charts, and assurance-checked output.
 - **Encrypted remote access** — pair the installable web app with Desktop and
   use Mixdog from a browser or phone over authenticated E2EE.
 - **Desktop coding app** — agent panes, Monaco editor, Git, terminals, file
@@ -180,7 +191,6 @@ stdout; diagnostics remain on stderr.
 /mcp          manage MCP servers and tools
 /skills       choose a skill for the next request
 /plugins      manage local plugin integrations
-/hooks        manage before-tool hooks and events
 /setting      open runtime settings
 /profile      set your title and response language
 /update       check for updates
@@ -202,14 +212,20 @@ Mixdog Desktop runs the same agent runtime as the CLI:
 - Git staging, commits, branches, and generated commit messages
 - File explorer with previews, thumbnails, search, and drag-and-drop
 - Integrated terminal tabs using the local system shell
+- Browser Use pane with agent control and Chromium profile import
+- Computer Use on Windows with guarded native input
+- Office document authoring with rendered previews
 - Image and video generation Studio with a persistent local gallery
 - Visual workflow, agent, schedule, and webhook editors
 - Voice dictation with an optional local transcription runtime
+- Extensions hub with guided setup for Git, Memory, Browser Use, Computer
+  Use, Office, and voice
 - Provider setup, usage, git identity, and remote pairing settings
 
 The paired remote web app is installable on desktop and mobile browsers. It
 uses an authenticated end-to-end encrypted connection before session state,
-terminal data, files, or operation requests cross the relay.
+terminal data, files, or operation requests cross the relay, and adds mobile
+share-target intake, push notifications, and remote Browser Use.
 
 For desktop development:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mixdog",
-  "version": "0.9.157",
+  "version": "0.9.159",
   "private": false,
   "type": "module",
   "description": "Standalone mixdog coding-agent CLI/TUI workspace.",
@@ -32,6 +32,10 @@
     "!scripts/recall-bench-*.txt",
     "!scripts/*-test.mjs",
     "!scripts/*.test.mjs",
+    "!scripts/tool-contracts/",
+    "!scripts/provider-toolcall/",
+    "!scripts/session-transport/",
+    "!scripts/shellhardening/",
     "!scripts/*-smoke.mjs",
     "!scripts/smoke-loop*.mjs",
     "!scripts/*-bench.mjs",
@@ -54,25 +58,25 @@
     "smoke:all": "npm run smoke && npm run smoke:boot && npm run smoke:patch && npm run smoke:compact",
     "smoke:boot": "node scripts/boot-smoke.mjs",
     "smoke:compact": "node scripts/compact-smoke.mjs",
-    "test:compact": "node --test scripts/suite-compact-test.mjs src/runtime/agent/orchestrator/session/token-estimate.test.mjs",
+    "test:compact": "node --test scripts/suite-compact-test.mjs src/runtime/agent/orchestrator/session/compact-fresh-context.test.mjs src/runtime/agent/orchestrator/session/compaction-read-reset.test.mjs src/runtime/agent/orchestrator/session/token-estimate.test.mjs",
     "test:markdown": "node --test scripts/markdown-surface-test.mjs",
     "smoke:loop": "node scripts/smoke-loop.mjs",
     "smoke:loop:final": "node scripts/smoke-loop-report.mjs --require-complete --min-elapsed 5h --min-iterations 400 --max-gap 60s --max-smoke-ms 10000 --max-avg-smoke-ms 8500 --max-step-ms smoke.mjs=4000 --max-step-ms boot-smoke.mjs=8000 --max-rss-mb 140 --max-rss-growth-mb 50",
     "smoke:loop:report": "node scripts/smoke-loop-report.mjs",
     "build:spawn:test": "cargo build --locked --manifest-path native/mixdog-spawn/Cargo.toml",
-    "test:tool-contracts": "npm run build:spawn:test && node scripts/tool-smoke.mjs",
+    "test:tool-contracts": "npm run build:spawn:test && node --test --test-force-exit \"scripts/tool-contracts/*.test.mjs\"",
     "smoke:patch": "node scripts/apply-patch-edit-smoke.mjs",
-    "test:shellhardening": "npm run build:spawn:test && node --test scripts/suite-shellhardening-test.mjs scripts/suite-shell-direct-exe-test.mjs",
-    "test:providers": "node --test scripts/provider-toolcall-test.mjs scripts/provider-contract-test.mjs scripts/provider-stream-outcome-test.mjs scripts/cursor-provider-test.mjs src/runtime/agent/orchestrator/providers/openai-codex-identity.test.mjs src/runtime/agent/orchestrator/providers/openai-ws-delta-parity.test.mjs",
+    "test:shellhardening": "npm run build:spawn:test && node --test \"scripts/shellhardening/*.test.mjs\" scripts/suite-shell-direct-exe-test.mjs",
+    "test:providers": "node --test \"scripts/provider-toolcall/*.test.mjs\" scripts/provider-contract-test.mjs scripts/provider-stream-outcome-test.mjs scripts/cursor-provider-test.mjs src/runtime/agent/orchestrator/providers/openai-codex-identity.test.mjs src/runtime/agent/orchestrator/providers/openai-ws-delta-parity.test.mjs",
     "test:embedding-runtime:warmup": "node scripts/verify-embedding-runtime.mjs --warmup",
     "test:release-assets": "node --check scripts/verify-release-assets.mjs && node --check scripts/release-gate-test.mjs && node --check scripts/prepare-native-assets.mjs && node --test scripts/release-gate-test.mjs scripts/prepare-native-assets-test.mjs scripts/generate-voice-runtime-manifest.test.mjs",
     "test:native-edit-wire": "node scripts/native-edit-wire-test.mjs",
     "test:release-critical": "npm run test:release-assets && npm run smoke:patch && npm run test:providers",
     "test:spec-advisory": "node scripts/advisory-spec-test.mjs && npm run test:spec-advisory --prefix apps/desktop",
-    "test:session-transport": "node --test scripts/session-transport-test.mjs scripts/daemon-bootstrap-test.mjs src/runtime/shared/child-spawn-remote.test.mjs src/standalone/session-runtime-agent-control-client.test.mjs src/standalone/session-runtime-dispatch-cancel.test.mjs src/standalone/session-runtime-host-factory.test.mjs src/standalone/session-runtime-inline-host.test.mjs src/standalone/session-runtime-provider-cooldown.test.mjs src/standalone/session-runtime-shard-host.test.mjs src/standalone/session-runtime-shard-router.test.mjs",
-    "test:session": "node --test scripts/runtime-turn-contract-test.mjs scripts/session-save-fault-store-test.mjs scripts/turn-review-revert-test.mjs src/runtime/shared/tool-surface.test.mjs src/runtime/bridge-clients.test.mjs",
+    "test:session-transport": "node --test \"scripts/session-transport/*.test.mjs\" scripts/daemon-bootstrap-test.mjs src/runtime/shared/child-spawn-remote.test.mjs src/standalone/session-runtime-agent-control-client.test.mjs src/standalone/session-runtime-dispatch-cancel.test.mjs src/standalone/session-runtime-host-factory.test.mjs src/standalone/session-runtime-inline-host.test.mjs src/standalone/session-runtime-provider-cooldown.test.mjs src/standalone/session-runtime-shard-host.test.mjs src/standalone/session-runtime-shard-router.test.mjs",
+    "test:session": "node --test scripts/runtime-turn-contract-test.mjs scripts/session-save-fault-store-test.mjs scripts/turn-review-revert-test.mjs src/runtime/shared/tool-surface.test.mjs src/runtime/bridge-clients.test.mjs src/runtime/computer-bridge/client-semantic-error.test.mjs src/runtime/agent/orchestrator/session/tool-envelope.test.mjs src/runtime/agent/orchestrator/session/eager-dispatch.test.mjs",
     "test:quick": "npm run test:session && npm run test:session-transport && npm run test:desktop-main --prefix apps/desktop",
-    "test:office": "node --test scripts/run-office-live-tests.test.mjs src/runtime/office/office-runtime.test.mjs src/runtime/office/office-assurance.test.mjs src/runtime/office/portable-authoring.test.mjs && npm run test:office-approval --prefix apps/desktop",
+    "test:office": "node --test scripts/run-office-live-tests.test.mjs src/runtime/office/office-runtime.test.mjs src/runtime/office/office-assurance.test.mjs src/runtime/office/portable-authoring.test.mjs src/runtime/office/slide-quality.test.mjs && npm run test:office-approval --prefix apps/desktop",
     "test:office:live": "node scripts/run-office-live-tests.mjs",
     "test:media": "node --test src/runtime/media/store.test.mjs src/runtime/media/renditions.test.mjs src/runtime/media/adapters/codex-image.test.mjs",
     "failures": "node scripts/tool-failures.mjs",
@@ -91,6 +95,7 @@
     "bench:recall": "node scripts/recall-bench.mjs",
     "bench:recall:strict": "node scripts/recall-bench.mjs --strict",
     "test:recall-validator": "node --test scripts/recall-bench-test.mjs",
+    "test:suite-health": "node --test scripts/suite-health.test.mjs",
     "bench:tui-load": "node scripts/tui-runtime-load-bench.mjs",
     "bench:output-style": "node scripts/output-style-bench.mjs",
     "bench:session-context": "node scripts/session-context-bench.mjs",

package/scripts/build-runtime-linux.sh

@@ -78,6 +78,7 @@ else
   make -j"$(nproc)"
   make install
   make -C contrib/pgcrypto install
+  make -C contrib/pg_trgm install
 fi
 
 PG_CONFIG="$STAGE_DIR/bin/pg_config"
@@ -255,6 +256,10 @@ for attempt in 1 2 3; do
     echo "  vector extension version: $EXTV"
     echo "  distance query result:    $DIST"
     [[ "$EXTV" == "$PGVECTOR_VERSION" ]] || { echo "FAIL: extversion=$EXTV expected=$PGVECTOR_VERSION"; exit 1; }
+    "${RUN_AS[@]}" "$RUNTIME_DIR/bin/psql" -h 127.0.0.1 -p "$SMOKE_PORT" -U postgres -d postgres -c "CREATE EXTENSION pg_trgm;" > /dev/null
+    SIM="$("${RUN_AS[@]}" "$RUNTIME_DIR/bin/psql" -h 127.0.0.1 -p "$SMOKE_PORT" -U postgres -d postgres -tAc "SELECT similarity('mixdog','mixdig') > 0;")"
+    echo "  pg_trgm similarity probe: $SIM"
+    [[ "$SIM" == "t" ]] || { echo "FAIL: pg_trgm similarity=$SIM"; exit 1; }
     "${RUN_AS[@]}" "$RUNTIME_DIR/bin/pg_ctl" -D "$SMOKE_DATA" -m fast stop > /dev/null
   )
   attempt_rc=$?
@@ -323,6 +328,8 @@ for attempt in 1 2 3; do
       "$EXTRACT_DIR/bin/psql" -h 127.0.0.1 -p "$EXTRACT_PORT" -U postgres -d postgres -tAc \
       "SELECT extversion FROM pg_extension WHERE extname='vector';")"
     [[ "$EXTV2" == "$PGVECTOR_VERSION" ]] || { echo "FAIL: extracted-smoke extversion=$EXTV2"; exit 1; }
+    "${RUN_AS[@]}" env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" \
+      "$EXTRACT_DIR/bin/psql" -h 127.0.0.1 -p "$EXTRACT_PORT" -U postgres -d postgres -c "CREATE EXTENSION pg_trgm;" > /dev/null
     "${RUN_AS[@]}" env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" \
       "$EXTRACT_DIR/bin/pg_ctl" -D "$EXTRACT_DATA" -m fast stop > /dev/null
   )

package/scripts/build-runtime-macos.sh

@@ -73,6 +73,7 @@ else
   make -j"$(sysctl -n hw.logicalcpu)"
   make install
   make -C contrib/pgcrypto install
+  make -C contrib/pg_trgm install
 fi
 
 PG_CONFIG="$STAGE_DIR/bin/pg_config"
@@ -165,6 +166,10 @@ DIST="$("$RUNTIME_DIR/bin/psql" -h 127.0.0.1 -p "$SMOKE_PORT" -U postgres -d pos
 echo "  vector extension version: $EXTV"
 echo "  distance query result:    $DIST"
 [[ "$EXTV" == "$PGVECTOR_VERSION" ]] || { echo "FAIL: extversion=$EXTV expected=$PGVECTOR_VERSION"; exit 1; }
+"$RUNTIME_DIR/bin/psql" -h 127.0.0.1 -p "$SMOKE_PORT" -U postgres -d postgres -c "CREATE EXTENSION pg_trgm;" > /dev/null
+SIM="$("$RUNTIME_DIR/bin/psql" -h 127.0.0.1 -p "$SMOKE_PORT" -U postgres -d postgres -tAc "SELECT similarity('mixdog','mixdig') > 0;")"
+echo "  pg_trgm similarity probe: $SIM"
+[[ "$SIM" == "t" ]] || { echo "FAIL: pg_trgm similarity=$SIM"; exit 1; }
 "$RUNTIME_DIR/bin/pg_ctl" -D "$SMOKE_DATA" -m fast stop > /dev/null
 trap - EXIT
 rm -rf "$SMOKE_DATA"
@@ -207,6 +212,8 @@ env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" \
   "$EXTRACT_DIR/bin/psql" -h 127.0.0.1 -p "$EXTRACT_PORT" -U postgres -d postgres -c "CREATE EXTENSION vector;" > /dev/null
 EXTV2="$(env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" "$EXTRACT_DIR/bin/psql" -h 127.0.0.1 -p "$EXTRACT_PORT" -U postgres -d postgres -tAc "SELECT extversion FROM pg_extension WHERE extname='vector';")"
 [[ "$EXTV2" == "$PGVECTOR_VERSION" ]] || { echo "FAIL: extracted-smoke extversion=$EXTV2"; exit 1; }
+env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" \
+  "$EXTRACT_DIR/bin/psql" -h 127.0.0.1 -p "$EXTRACT_PORT" -U postgres -d postgres -c "CREATE EXTENSION pg_trgm;" > /dev/null
 env -i HOME="$EXTRACT_DIR" PATH="/usr/bin:/bin" \
   "$EXTRACT_DIR/bin/pg_ctl" -D "$EXTRACT_DATA" -m fast stop > /dev/null
 trap - EXIT

package/scripts/lib/stage-postgres-runtime-windows.ps1

@@ -64,6 +64,7 @@ function Stage-MixdogWindowsPgRuntime {
     )
     $ModuleSources = [ordered]@{
         'dict_snowball.dll' = Join-Path $PgRoot 'lib\dict_snowball.dll'
+        'pg_trgm.dll'       = Join-Path $PgRoot 'lib\pg_trgm.dll'
         'plpgsql.dll'       = Join-Path $PgRoot 'lib\plpgsql.dll'
         'vector.dll'        = $VectorDll
     }
@@ -137,6 +138,15 @@ function Stage-MixdogWindowsPgRuntime {
     }
     $PlpgsqlAssets | Copy-Item -Destination $RuntimeExtensionDir -Force
 
+    # pg_trgm ships with the runtime: memory recall's hybrid trigram CTEs and
+    # transcript inserts depend on it (runtime-v0.4.0 omitted it and every
+    # memory write/recall failed with 'could not access file "$libdir/pg_trgm"').
+    $PgTrgmAssets = @(Get-ChildItem -Path "$SourceShare\extension\pg_trgm*" -File)
+    if ($PgTrgmAssets.Count -eq 0) {
+        throw "Required pg_trgm extension assets not found under $SourceShare\extension"
+    }
+    $PgTrgmAssets | Copy-Item -Destination $RuntimeExtensionDir -Force
+
     $RuntimeBytes = (
         Get-ChildItem -LiteralPath $RuntimeDir -File -Recurse |
             Measure-Object -Property Length -Sum
@@ -174,7 +184,7 @@ function Assert-MixdogWindowsPgRuntime {
         throw "Unexpected PostgreSQL executable set: $($ActualExecutables -join ', ')"
     }
 
-    $ExpectedModules = @('dict_snowball.dll', 'plpgsql.dll', 'vector.dll')
+    $ExpectedModules = @('dict_snowball.dll', 'pg_trgm.dll', 'plpgsql.dll', 'vector.dll')
     $ActualModules = @(
         Get-ChildItem -LiteralPath "$RuntimeDir\lib" -File -Filter '*.dll' |
             ForEach-Object Name |
@@ -195,7 +205,7 @@ function Assert-MixdogWindowsPgRuntime {
 
     $UnexpectedExtensionAssets = @(
         Get-ChildItem -LiteralPath "$RuntimeDir\share\extension" -Force |
-            Where-Object { $_.Name -notlike 'plpgsql*' -and $_.Name -notlike 'vector*' }
+            Where-Object { $_.Name -notlike 'pg_trgm*' -and $_.Name -notlike 'plpgsql*' -and $_.Name -notlike 'vector*' }
     )
     if ($UnexpectedExtensionAssets.Count -gt 0) {
         throw "Unexpected extension assets leaked into runtime: $($UnexpectedExtensionAssets.Count)"
@@ -205,6 +215,7 @@ function Assert-MixdogWindowsPgRuntime {
         "$RuntimeDir\share\extension\vector.control",
         "$RuntimeDir\share\extension\vector--$PgVectorVersion.sql",
         "$RuntimeDir\share\extension\plpgsql.control",
+        "$RuntimeDir\share\extension\pg_trgm.control",
         "$RuntimeDir\share\postgres.bki"
     )
     foreach ($Path in $RequiredPaths) {

package/scripts/office-design-audit.mjs

@@ -0,0 +1,43 @@
+import { readFile } from 'node:fs/promises';
+import { basename, resolve } from 'node:path';
+import { pathToFileURL } from 'node:url';
+
+import { reviewRenderedOfficePages } from '../src/runtime/office/assurance.mjs';
+
+function argumentValue(name) {
+  const index = process.argv.indexOf(name);
+  return index >= 0 ? process.argv[index + 1] : '';
+}
+
+function pageFromPath(path, fallback) {
+  const match = basename(path).match(/(?:page|slide)[-_ ]?(\d+)/i);
+  return match ? Number(match[1]) : fallback;
+}
+
+export async function auditOfficeDesignImages(paths, {
+  format = 'pptx',
+} = {}) {
+  const images = await Promise.all(paths.map(async (path, index) => {
+    const absolute = resolve(path);
+    const data = await readFile(absolute);
+    return {
+      page: pageFromPath(absolute, index + 1),
+      path: absolute,
+      mimeType: 'image/png',
+      data: data.toString('base64'),
+    };
+  }));
+  images.sort((left, right) => left.page - right.page);
+  return reviewRenderedOfficePages(images, { format });
+}
+
+if (process.argv[1] && import.meta.url === pathToFileURL(process.argv[1]).href) {
+  const format = argumentValue('--format') || 'pptx';
+  const paths = process.argv.slice(2).filter((entry, index, values) => (
+    entry !== '--format' && values[index - 1] !== '--format'
+  ));
+  if (!paths.length) throw new Error('Pass rendered page image paths to audit.');
+  const report = await auditOfficeDesignImages(paths, { format });
+  process.stdout.write(`${JSON.stringify(report, null, 2)}\n`);
+  if (!report.ok) process.exitCode = 1;
+}

package/scripts/release-paths.mjs

@@ -0,0 +1,108 @@
+#!/usr/bin/env node
+// Single source of truth for CI/CD path selection.
+//
+// Consumers:
+//   - .github/workflows/release-gate.yml  → `node scripts/release-paths.mjs desktop-regex|runtime-regex`
+//     (grep -E patterns selecting which incremental gates a change requires)
+//   - .github/workflows/deploy.yml        → `import { RELEASE_CRITICAL_PATHS }`
+//     (git diff paths whose change invalidates the previous release's
+//     critical verification)
+//
+// Adding or renaming a release-relevant script means editing THIS file only;
+// the workflows derive their patterns from it at run time.
+
+export const PACKAGE_MANIFESTS = ['package.json', 'package-lock.json'];
+
+// This module governs gate selection itself, so changing it must re-run
+// every gate it feeds.
+const GATE_SELECTION_SOURCES = ['scripts/release-paths.mjs'];
+
+// Prefixes whose changes require the desktop bundle gate.
+export const DESKTOP_GATE_PREFIXES = ['apps/desktop/', 'src/', 'vendor/', 'LICENSES/'];
+export const DESKTOP_GATE_FILES = ['README.md', 'NOTICE.md', ...PACKAGE_MANIFESTS];
+// Desktop packaging/postinstall scripts; `.test` companions ride along.
+export const DESKTOP_GATE_SCRIPT_BASES = [
+  'scripts/prune-embedding-runtime',
+  'scripts/native-binary-arch',
+  'scripts/native-tool-download',
+  'scripts/runtime-dependency-cache-key',
+];
+
+// Prefixes whose changes require the runtime gate (the release-critical lane
+// plus the tool-contract suites the gate also executes).
+export const RUNTIME_GATE_PREFIXES = [
+  '.github/workflows/',
+  'native/',
+  'scripts/tool-contracts/',
+  'src/runtime/agent/orchestrator/',
+  'src/runtime/shared/',
+];
+
+// Everything `npm run test:release-critical` executes
+// (test:release-assets + smoke:patch + test:providers).
+export const RELEASE_CRITICAL_SCRIPTS = [
+  'scripts/verify-release-assets.mjs',
+  'scripts/release-gate-test.mjs',
+  'scripts/prepare-native-assets.mjs',
+  'scripts/prepare-native-assets-test.mjs',
+  'scripts/generate-voice-runtime-manifest.mjs',
+  'scripts/generate-voice-runtime-manifest.test.mjs',
+  'scripts/apply-patch-edit-smoke.mjs',
+  'scripts/provider-contract-test.mjs',
+  'scripts/provider-stream-outcome-test.mjs',
+  'scripts/cursor-provider-test.mjs',
+];
+
+// Directory-shaped suites the critical lane executes (test:providers runs
+// scripts/provider-toolcall/*.test.mjs): deploy consumes these as git
+// pathspecs, the gate regex as path prefixes.
+export const RELEASE_CRITICAL_SUITE_DIRS = ['scripts/provider-toolcall'];
+
+// Gate-only suite files the runtime job executes beyond the critical lane
+// (test:compact); editing one must re-run the gate that executes it.
+export const RUNTIME_GATE_SUITE_FILES = ['scripts/suite-compact-test.mjs'];
+
+// deploy.yml plan: paths whose change since the published release tag marks
+// the critical lane as unverified.
+export const RELEASE_CRITICAL_PATHS = [
+  'native',
+  'src/runtime/agent/orchestrator',
+  'src/runtime/shared',
+  ...RELEASE_CRITICAL_SCRIPTS,
+  ...RELEASE_CRITICAL_SUITE_DIRS,
+  ...PACKAGE_MANIFESTS,
+];
+
+const escapeRegex = (text) => text.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+
+export function desktopGateRegex() {
+  return `^(${[
+    ...DESKTOP_GATE_PREFIXES.map(escapeRegex),
+    ...DESKTOP_GATE_FILES.map((file) => `${escapeRegex(file)}$`),
+    ...DESKTOP_GATE_SCRIPT_BASES.map((base) => `${escapeRegex(base)}(\\.test)?\\.mjs$`),
+    ...GATE_SELECTION_SOURCES.map((file) => `${escapeRegex(file)}$`),
+  ].join('|')})`;
+}
+
+export function runtimeGateRegex() {
+  return `^(${[
+    ...RUNTIME_GATE_PREFIXES.map(escapeRegex),
+    ...RELEASE_CRITICAL_SCRIPTS.map((file) => `${escapeRegex(file)}$`),
+    ...RELEASE_CRITICAL_SUITE_DIRS.map((dir) => `${escapeRegex(dir)}/`),
+    ...RUNTIME_GATE_SUITE_FILES.map((file) => `${escapeRegex(file)}$`),
+    ...DESKTOP_GATE_SCRIPT_BASES.map((base) => `${escapeRegex(base)}(\\.test)?\\.mjs$`),
+    ...PACKAGE_MANIFESTS.map((file) => `${escapeRegex(file)}$`),
+    ...GATE_SELECTION_SOURCES.map((file) => `${escapeRegex(file)}$`),
+  ].join('|')})`;
+}
+
+const mode = process.argv[2];
+if (mode) {
+  if (mode === 'desktop-regex') process.stdout.write(`${desktopGateRegex()}\n`);
+  else if (mode === 'runtime-regex') process.stdout.write(`${runtimeGateRegex()}\n`);
+  else if (mode === 'critical-paths') process.stdout.write(`${RELEASE_CRITICAL_PATHS.join('\n')}\n`);
+  else {
+    process.stderr.write(`unknown mode: ${mode} (expected desktop-regex|runtime-regex|critical-paths)\n`);
+    process.exit(1);
+  }
+}

package/scripts/smoke-runtime-negative.sh

@@ -9,7 +9,7 @@
 
 set -euo pipefail
 
-TAG="${TAG:-runtime-v0.4.0}"
+TAG="${TAG:-runtime-v0.4.1}"
 OS="${OS:-$(uname -s | tr '[:upper:]' '[:lower:]')}"
 ARCH="${ARCH:-$(uname -m | sed 's/x86_64/x64/')}"
 PG_VER="16.4"
@@ -76,7 +76,8 @@ echo "==> ldd vector.so (under env -i):"
 "${RUN_AS[@]}" env -i HOME="$WORK/fresh" PATH=/usr/bin:/bin "$PG_BIN/psql" -h 127.0.0.1 -p "$PORT" -U postgres -d postgres -c "CREATE EXTENSION vector;" > /dev/null
 EXTV="$("${RUN_AS[@]}" env -i HOME="$WORK/fresh" PATH=/usr/bin:/bin "$PG_BIN/psql" -h 127.0.0.1 -p "$PORT" -U postgres -d postgres -tAc "SELECT extversion FROM pg_extension WHERE extname='vector';")"
 [[ "$EXTV" == "$PGVECTOR_VER" ]] || { echo "FAIL: extversion=$EXTV"; exit 1; }
-echo "  PASS: fresh-extract boot + vector extension"
+"${RUN_AS[@]}" env -i HOME="$WORK/fresh" PATH=/usr/bin:/bin "$PG_BIN/psql" -h 127.0.0.1 -p "$PORT" -U postgres -d postgres -c "CREATE EXTENSION pg_trgm;" > /dev/null
+echo "  PASS: fresh-extract boot + vector + pg_trgm extensions"
 
 echo "==> Test 5: same tarball extracted twice → second initdb refuses (data-dir already initialized)"
 mkdir -p "$WORK/twice"

package/scripts/suite-health.mjs

@@ -0,0 +1,102 @@
+#!/usr/bin/env node
+// Weekly opt-out sweep: run every test:*/smoke:* npm script in the repository
+// so no locally-registered suite can rot silently (docs/testing.md, "A
+// contract suite runs in CI or it drifts"). New scripts join automatically;
+// exclusions are explicit, reasoned, and fail closed when they go stale.
+import { readFileSync, writeFileSync } from 'node:fs';
+import { join } from 'node:path';
+import { pathToFileURL } from 'node:url';
+import { spawnSync } from 'node:child_process';
+
+export const WORKSPACES = [
+  {
+    dir: '.',
+    exclusions: {
+      'smoke:all': 'aggregate; its components run individually',
+      'smoke:loop': 'long-running local endurance harness',
+      'smoke:loop:final': 'endurance-harness gate, needs hours of loop data',
+      'smoke:loop:report': 'report reader, not a suite',
+      'test:release-critical': 'aggregate; the release gate owns it',
+      'test:quick': 'aggregate; its components run individually',
+      'test:office:live': 'drives real Office COM on a desktop machine',
+      'test:embedding-runtime:warmup': 'downloads the embedding runtime',
+      'test:spec-advisory': 'advisory by design (MIXDOG_TEST_ADVISORY)',
+    },
+  },
+  {
+    dir: 'apps/desktop',
+    exclusions: {
+      'test:renderer:full': 'aggregate; its components run individually',
+      'test:fast-direct': 'dev-only deploy tool; real deploys are its verification',
+      'test:office-approval': 'runs through the root test:office chain',
+      'test:spec-advisory': 'advisory by design (MIXDOG_TEST_ADVISORY)',
+      'test:browser-host:integration': 'live desktop integration harness',
+      'test:browser-profile-import:integration': 'live desktop integration harness',
+      'test:computer-host:integration': 'live desktop integration harness',
+      'test:computer-host:scenarios': 'live desktop scenario harness',
+      'test:computer-host:repeats': 'live desktop scenario harness',
+    },
+  },
+];
+
+export function selectSuiteScripts(scripts, exclusions) {
+  const names = Object.keys(scripts).filter((name) => /^(test|smoke)(:|$)/.test(name));
+  const stale = Object.keys(exclusions).filter((name) => !names.includes(name));
+  if (stale.length) {
+    throw new Error(`stale exclusions (script no longer exists): ${stale.join(', ')}`);
+  }
+  return names.filter((name) => !(name in exclusions));
+}
+
+function runSweep() {
+  const failures = [];
+  const reportLines = [];
+  for (const ws of WORKSPACES) {
+    const pkg = JSON.parse(readFileSync(join(ws.dir, 'package.json'), 'utf8'));
+    const names = selectSuiteScripts(pkg.scripts || {}, ws.exclusions);
+    for (const name of names) {
+      const label = ws.dir === '.' ? name : `${ws.dir} ${name}`;
+      const started = Date.now();
+      const args = ['run', name, ...(ws.dir === '.' ? [] : ['--prefix', ws.dir])];
+      const child = spawnSync(process.platform === 'win32' ? 'npm.cmd' : 'npm', args, {
+        encoding: 'utf8',
+        shell: process.platform === 'win32',
+        maxBuffer: 64 * 1024 * 1024,
+      });
+      const seconds = ((Date.now() - started) / 1000).toFixed(1);
+      if (child.status === 0) {
+        reportLines.push(`- ${label}: OK (${seconds}s)`);
+        console.log(`${label}: OK (${seconds}s)`);
+      } else {
+        const tail = `${child.stdout || ''}\n${child.stderr || ''}`
+          .trim().split('\n').slice(-25).join('\n');
+        failures.push(`- ${label}: FAILED (${seconds}s)\n\n\`\`\`\n${tail}\n\`\`\``);
+        reportLines.push(`- ${label}: FAILED (${seconds}s)`);
+        console.error(`${label}: FAILED (${seconds}s)\n${tail}`);
+      }
+    }
+  }
+  console.log(`\n${reportLines.join('\n')}`);
+  if (failures.length > 0) {
+    const runUrl = process.env.GITHUB_RUN_ID
+      ? `${process.env.GITHUB_SERVER_URL}/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}`
+      : '(local run)';
+    const report = [
+      '# Weekly suite health sweep failed',
+      '',
+      `Checked: ${new Date().toISOString()}`,
+      `Workflow run: ${runUrl}`,
+      '',
+      ...reportLines,
+      '',
+      ...failures,
+      '',
+    ].join('\n');
+    writeFileSync('suite-health-report.md', report);
+    process.exitCode = 1;
+  }
+}
+
+if (process.argv[1] && import.meta.url === pathToFileURL(process.argv[1]).href) {
+  runSweep();
+}

package/src/defaults/skills/setup/SKILL.md

@@ -18,7 +18,7 @@ description: Use this skill only to inspect or modify a Mixdog user's persisted
 - 프로젝트 config: `<cwd>/.mixdog/hooks.json`
 - TUI 명령 목록: `src/tui/app/slash-commands.mjs`
 - TUI 설정 허브: `/setting` (별칭 `/settings`, `/config`) → `src/tui/app/settings-picker.mjs`
-- Desktop 설정: `Ctrl+,` → General / Context / Output style / Providers / Git / Skills / MCP / Plugins / Hooks / Connection / System / Shortcuts / About
+- Desktop 설정: `Ctrl+,` → General / Context / Output style / Providers / Git / Skills / MCP / Plugins / Connection / System / Shortcuts / About
 - Desktop rail: **Projects / Workflows / Schedules / Webhooks / Utilities**
 
 사용자 경로와 지원 여부는 **현재 Desktop 구현을 최신 기준**으로 판정한다. TUI는 Desktop에 없는 고급 경로를 보완할 때만 사용하며, 저장만 왕복되고 실제 소비되지 않는 key는 사용자 옵션으로 설명하지 않는다.
@@ -219,7 +219,7 @@ API 키·토큰·OAuth 자격은 config에 쓰지 않는다.
 
 1. **확인**: `/setting → Auto-compact`, Desktop **Settings → Context**, 디스크 `agent.compaction`.
 2. **변경**: Auto-compact toggle → `setCompactionSettings({ auto })`.
-3. **고정 계약**: Compact type은 `Fast-track (fixed)`이며 UI 변경 불가. memory recap이 꺼져도 recall-fasttrack은 유지된다.
+3. **고정 계약**: 모든 session은 하나의 fresh-context Compact를 사용한다. Main은 Memory handoff, Agent는 session-local handoff를 사용하며 type 선택 UI는 없다.
 
 ### Memory / Recap / Core Memory
 
@@ -260,7 +260,7 @@ TUI·Desktop interval 편집 UI는 없다.
 
 ---
 
-## 4. MCP·Skills·Plugins·Hooks
+## 4. MCP·Skills·Plugins·Hooks(내부)
 
 ### MCP 추가
 
@@ -319,14 +319,13 @@ UI 생성 액션은 없다.
 
 Plugin toggle은 해당 Plugin의 Skills와 MCP를 전역으로 함께 활성화/비활성화한다. 개별 MCP와 Skill은 각 화면에서도 관리할 수 있다.
 
-### Hooks
+### Hooks (사용자 설정 UI 없음)
 
-1. **확인**: `/hooks` 또는 Desktop **Settings → Hooks**, `hooksStatus()`.
-2. **UI 범위**: 기존 rule의 Enable/Disable만 지원한다. UI에서 add/delete/edit은 하지 않는다.
-3. **추가·수정·삭제**: hook config 파일을 편집하거나 runtime `addHookRule()` / `deleteHookRule()`을 사용한다.
-4. **검색 경로**: `MIXDOG_HOOKS_FILE` → 프로젝트 `.mixdog/hooks.json`·`.mixdog/hooks/hooks.json` → 글로벌 `<mixdogData>/hooks.json`·`hooks/hooks.json` → plugin hooks.
-5. **형식**: 표준 `{ "hooks": { "<Event>": [{ "matcher": "...", "hooks": [...] }] } }`; 레거시 before-tool 배열도 읽는다.
-6. **검증**: mtime 기반 자동 reload 후 Hooks 목록과 대상 event 발동을 확인한다.
+1. **범위**: hook은 plugin·내부 연동 전용이다. Desktop 설정 화면과 `/hooks` 슬래시 명령은 제공하지 않으므로 사용자 옵션으로 안내하지 않는다.
+2. **구성**: hook config 파일을 직접 편집하거나 runtime `addHookRule()` / `setHookRuleEnabled()` / `deleteHookRule()`을 사용한다.
+3. **검색 경로**: `MIXDOG_HOOKS_FILE` → 프로젝트 `.mixdog/hooks.json`·`.mixdog/hooks/hooks.json` → 글로벌 `<mixdogData>/hooks.json`·`hooks/hooks.json` → plugin hooks.
+4. **형식**: 표준 `{ "hooks": { "<Event>": [{ "matcher": "...", "hooks": [...] }] } }`; 레거시 before-tool 배열도 읽는다.
+5. **검증**: mtime 기반 자동 reload 후 `/doctor`의 hooks 행과 대상 event 발동을 확인한다.
 
 ---
 
@@ -415,7 +414,7 @@ Desktop과 현재 runtime에서 소비되지 않는 아래 key는 사용자 옵
 | `channels.webhook.ngrokDomain`, `channels.webhook.respectQuiet` | 제거; endpoint URL은 Desktop Webhooks가 발급 |
 | `agent.outputStyle` | 제거 (이관 없음; 루트 `outputStyle`만 유효) |
 
-다음 항목은 레거시처럼 보여도 현행이므로 유지한다: 루트 `outputStyle`, `agent.shell`, `agent.modules`, compaction의 `type/compactType`과 recall tuning fields.
+다음 항목은 레거시처럼 보여도 현행이므로 유지한다: 루트 `outputStyle`, `agent.shell`, `agent.modules`, compaction의 `auto`, `summaryModel`, `timeoutMs`, `memoryTimeoutMs`, buffer/target fields.
 
 ---
 

package/src/runtime/agent/orchestrator/agent-runtime/session-builder.mjs

@@ -23,22 +23,22 @@ import { resolveAgentSessionPermission } from '../internal-agents.mjs';
 import { loadConfig } from '../config.mjs';
 import { AGENT_OWNER } from '../agent-owner.mjs';
 
-import {
-    COMPACT_TYPE_SEMANTIC,
-    normalizeCompactType,
-} from '../session/compact.mjs';
-
 function normalizeAgentCompactionConfig(value = {}) {
     const raw = value && typeof value === 'object' ? value : {};
-    const compactType = normalizeCompactType(
-        raw.compactType ?? raw.compact_type ?? raw.type,
-        COMPACT_TYPE_SEMANTIC,
-    );
+    const next = { ...raw };
+    if (!next.summaryModel && raw.semanticModel) next.summaryModel = raw.semanticModel;
+    if (!next.memoryTimeoutMs && raw.recallMemoryTimeoutMs) next.memoryTimeoutMs = raw.recallMemoryTimeoutMs;
+    for (const key of [
+        'type', 'compactType', 'compact_type', 'semantic', 'semanticModel', 'prune', 'tailTurns',
+        'recallMemoryTimeoutMs', 'recallIngestLimit', 'recallChunkLimit', 'recallLimit',
+        'recallCycle1BatchSize', 'recallRowsPerSession', 'recallWindowSize',
+        'recallConcurrency', 'recallCycle1DeadlineMs',
+    ]) {
+        delete next[key];
+    }
     return {
-        ...raw,
+        ...next,
         auto: raw.auto !== false && raw.enabled !== false,
-        type: compactType,
-        compactType,
     };
 }