feat(sdk): support embedded hosts with durable task completion

.github/workflows/ci.yml

@@ -6,6 +6,9 @@ on:
     branches:
       - main
 
+permissions:
+  contents: read
+
 jobs:
   check:
     name: Check
@@ -29,8 +32,6 @@ jobs:
   test:
     name: Test (${{ matrix.os }}, node ${{ matrix.node }})
     runs-on: ${{ matrix.os }}
-    # Non-blocking for now (team decision 2026-10-11): failures stay visible but do not gate merges. Check stays blocking.
-    continue-on-error: true
     strategy:
       fail-fast: false
       matrix:
@@ -71,6 +72,13 @@ jobs:
             || python3 -m pip install --user --break-system-packages pyte
           python3 -c 'import pyte'
 
+      - name: Install real Windows ConPTY test driver
+        if: runner.os == 'Windows'
+        run: |
+          python --version
+          python -m pip install --only-binary=:all: pywinpty==3.0.5
+          python -c "from winpty import Backend, PtyProcess; print(Backend.ConPTY)"
+
       - name: Test (build + spec suite)
         env:
           MOSS_SKIP_PY_LAYOUT: ${{ runner.os == 'Windows' && '1' || '' }}
@@ -80,14 +88,29 @@ jobs:
         if: runner.os != 'Windows'
         run: npm run smoke
 
+      # Record this leg's outcome for the Test summary gate (see test-summary).
+      - name: Record leg result
+        if: always()
+        shell: bash
+        run: |
+          mkdir -p leg-result
+          echo "${{ matrix.os }} node ${{ matrix.node }}: ${{ job.status }}" > leg-result/result.txt
+          cat leg-result/result.txt
+
+      - name: Upload leg result
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: test-leg-${{ matrix.os }}-${{ matrix.node }}
+          path: leg-result/result.txt
+          retention-days: 7
+
   # Not part of the OS matrix. Ubuntu runners ship tmux and macOS runners ship
   # GNU screen, but those sessions do not meet the composer check (and Windows
   # has neither tool). This job installs the tools and opts in.
   real-terminals:
     name: Real terminals (Linux)
     runs-on: ubuntu-latest
-    # Non-blocking for now (team decision 2026-10-11): failures stay visible but do not gate merges. Check stays blocking.
-    continue-on-error: true
     timeout-minutes: 20
     steps:
       - name: Checkout
@@ -121,8 +144,6 @@ jobs:
   install-e2e:
     name: Install e2e (${{ matrix.os }}, npm 10)
     runs-on: ${{ matrix.os }}
-    # Non-blocking for now (team decision 2026-10-11): failures stay visible but do not gate merges. Check stays blocking.
-    continue-on-error: true
     timeout-minutes: 30
     strategy:
       fail-fast: false
@@ -158,8 +179,6 @@ jobs:
   windows-shell:
     name: Windows shell (pwsh 7)
     runs-on: windows-latest
-    # Non-blocking for now (team decision 2026-10-11): failures stay visible but do not gate merges. Check stays blocking.
-    continue-on-error: true
     timeout-minutes: 20
     steps:
       - name: Checkout
@@ -182,7 +201,82 @@ jobs:
         run: |
           $major = $PSVersionTable.PSVersion.Major
           if ($major -lt 7) { throw "runner pwsh $major is older than 7" }
-          node --input-type=module -e "import { resolveHostShell } from './dist/utils/host-shell.js'; const shell = resolveHostShell(); if (shell.kind !== 'pwsh') { console.error(shell); process.exit(1); } console.log(shell.kind + ' ' + shell.executable);"
+          node scripts/check-windows-host-shell.mjs
 
       - name: Host shell and write-sandbox specs
         run: node test/host-shell.spec.mjs && node test/shell-write-sandbox.spec.mjs
+
+  # Single required status for branch protection.
+  # Rule (Boss decision on #120, 2026-10-12): Test summary fails if Check fails
+  # or if any ubuntu-latest / macos-latest Test leg did not succeed (failure,
+  # cancelled, or no recorded result). windows-latest legs are REPORTED here
+  # but do NOT block yet: 25 spec files fail on Windows on main (tracked in
+  # specs/windows-25-tests.md, each to be fixed or explicitly skipped).
+  #
+  # Windows becomes blocking AUTOMATICALLY (deadline for #120 / Goal 4, O134):
+  # when the run date in UTC+8 (Asia/Shanghai) is after 2026-10-15, i.e. from
+  # 2026-10-16 00:00 UTC+8, Windows Test failures fail Test summary. No manual
+  # flip needed.
+  # Manual override via WINDOWS_BLOCKING:
+  #   'auto'  (default) date-based rule above
+  #   'true'  block on Windows now, regardless of date
+  #   'false' never block on Windows. Emergency only: requires a Boss decision
+  #           and a ticket number in the commit; do not extend silently.
+  test-summary:
+    name: Test summary
+    if: always()
+    needs:
+      - check
+      - test
+    runs-on: ubuntu-latest
+    env:
+      WINDOWS_BLOCKING: 'auto' # auto | true | false (see comment above)
+      WINDOWS_BLOCKING_AFTER: '20261015' # YYYYMMDD, UTC+8; blocking when run date > this
+    steps:
+      - name: Download leg results
+        uses: actions/download-artifact@v4
+        continue-on-error: true
+        with:
+          pattern: test-leg-*
+          path: legs
+
+      - name: Gate on Check + Test (Windows blocking after 2026-10-15 UTC+8)
+        env:
+          CHECK_RESULT: ${{ needs.check.result }}
+          TEST_RESULT: ${{ needs.test.result }}
+        run: |
+          fail=0
+          run_date=$(TZ=Asia/Shanghai date +%Y%m%d)
+          case "$WINDOWS_BLOCKING" in
+            true|false) win_block=$WINDOWS_BLOCKING; win_src=manual ;;
+            *) if [ "$run_date" -gt "$WINDOWS_BLOCKING_AFTER" ]; then win_block=true; else win_block=false; fi; win_src="auto(date=$run_date UTC+8, after=$WINDOWS_BLOCKING_AFTER)" ;;
+          esac
+          WINDOWS_BLOCKING=$win_block
+          echo "windows_blocking source: $win_src"
+          echo "check=$CHECK_RESULT test(matrix overall)=$TEST_RESULT windows_blocking=$WINDOWS_BLOCKING"
+          [ "$CHECK_RESULT" = success ] || { echo "::error::Check: $CHECK_RESULT"; fail=1; }
+          {
+            echo "## Test summary"
+            echo "Blocking: Check, ubuntu-latest, macos-latest. Windows blocking: $WINDOWS_BLOCKING ($win_src; auto-blocking after 2026-10-15 UTC+8)."
+            echo
+            echo "| Leg | Result | Blocking |"
+            echo "|---|---|---|"
+          } >> "$GITHUB_STEP_SUMMARY"
+          for os in ubuntu-latest macos-latest windows-latest; do
+            for node in 22.16.0 24; do
+              f="legs/test-leg-$os-$node/result.txt"
+              if [ -f "$f" ]; then r=$(awk '{print $NF}' "$f"); else r=missing; fi
+              blocking=yes
+              if [ "$os" = windows-latest ] && [ "$WINDOWS_BLOCKING" != true ]; then blocking=no; fi
+              echo "$os node $node: $r (blocking=$blocking)"
+              echo "| $os node $node | $r | $blocking |" >> "$GITHUB_STEP_SUMMARY"
+              if [ "$r" != success ]; then
+                if [ "$blocking" = yes ]; then
+                  echo "::error::Test ($os, node $node): $r"; fail=1
+                else
+                  echo "::warning::Test ($os, node $node): $r (non-blocking: $win_src)"
+                fi
+              fi
+            done
+          done
+          exit $fail

AGENTS.md

@@ -5,7 +5,7 @@
 ## 仓库身份
 
 Moss 是一个精简的跨平台 coding agent harness：TypeScript / ESM 单包仓库（根 `package.json`，
-包名 `moss`，bin 为 `dist/cli.js`），Node ≥ 22.16.0，运行于 Linux / macOS / Windows。
+包名 `@rdk-moss/agent`，bin 为 `dist/cli.js`），Node ≥ 22.16.0，运行于 Linux / macOS / Windows。
 
 核心能力（也是唯一应当存在的范围）：agent loop、工具框架（`src/tools/`）、上下文管理
 （`src/context/`）、provider（`src/provider/`）、安全（`src/safety/`）、会话