_smoke.yml

# Reusable: smoke test shipped release binaries on all platforms
name: Smoke

on:
  workflow_call:
    inputs:
      broad_platforms:
        description: 'Smoke the shipped binaries on the broad platform matrix (extra OS versions) instead of the core set'
        type: boolean
        default: false
permissions:
  contents: read

jobs:
  # Emit the platform matrices as JSON. The CORE set is the default (fast,
  # unchanged); the BROAD set adds extra free runners (additional OS versions)
  # that download the SAME shipped artifact for their goos/goarch and verify it
  # runs on a wider range of OS versions. No new artifacts are built — broad
  # legs reuse the exact binaries produced by _build.yml.
  setup-matrix:
    runs-on: ubuntu-latest
    timeout-minutes: 5
    outputs:
      unix: ${{ steps.set.outputs.unix }}
      windows: ${{ steps.set.outputs.windows }}
      portable: ${{ steps.set.outputs.portable }}
    steps:
      - name: Compute matrices
        id: set
        env:
          BROAD: ${{ inputs.broad_platforms }}
        run: |
          CORE_UNIX='[
            {"os":"ubuntu-latest","goos":"linux","goarch":"amd64"},
            {"os":"ubuntu-24.04-arm","goos":"linux","goarch":"arm64"},
            {"os":"macos-14","goos":"darwin","goarch":"arm64"},
            {"os":"macos-15-intel","goos":"darwin","goarch":"amd64"}
          ]'
          # Broad legs reuse existing goos/goarch artifacts on additional OS
          # versions to widen the run-anywhere signal without building new targets.
          # Broad legs are REQUIRED gates (no optional / continue-on-error):
          # every smoke leg must pass. No `optional` flags anywhere.
          #
          # NOTE: the *dynamic* linux binary links glibc 2.38+ and cannot run on
          # older distros by design — older-glibc coverage is the -portable (static)
          # binary's job, exercised green by the smoke-linux-portable broad legs
          # (ubuntu-22.04 / 22.04-arm). So the dynamic broad legs stay on
          # forward-compatible OSes only (macOS); running the dynamic binary on
          # ubuntu-22.04 would fail Phase 1 (glibc too old), not a real regression.
          BROAD_UNIX='[
            {"os":"macos-15","goos":"darwin","goarch":"arm64"}
          ]'
          CORE_WIN='[{"os":"windows-latest","arch":"amd64"}]'
          # windows-11-arm now runs the NATIVE arm64 binary (build-windows-arm64),
          # not the x86_64 binary under emulation — we smoke the artifact we ship.
          BROAD_WIN='[{"os":"windows-2025","arch":"amd64"},{"os":"windows-11-arm","arch":"arm64"}]'
          CORE_PORTABLE='[
            {"arch":"amd64","runner":"ubuntu-latest"},
            {"arch":"arm64","runner":"ubuntu-24.04-arm"}
          ]'
          BROAD_PORTABLE='[
            {"arch":"amd64","runner":"ubuntu-22.04"},
            {"arch":"arm64","runner":"ubuntu-22.04-arm"}
          ]'
          if [ "$BROAD" = "true" ]; then
            UNIX=$(jq -cn --argjson a "$CORE_UNIX" --argjson b "$BROAD_UNIX" '$a + $b')
            WIN=$(jq -cn --argjson a "$CORE_WIN" --argjson b "$BROAD_WIN" '$a + $b')
            PORTABLE=$(jq -cn --argjson a "$CORE_PORTABLE" --argjson b "$BROAD_PORTABLE" '$a + $b')
          else
            UNIX=$(jq -cn --argjson a "$CORE_UNIX" '$a')
            WIN=$(jq -cn --argjson a "$CORE_WIN" '$a')
            PORTABLE=$(jq -cn --argjson a "$CORE_PORTABLE" '$a')
          fi
          # One composition ships; the wrappers assert the embedded UI themselves
          # (SMOKE_REQUIRE_UI in scripts/ci/smoke-artifact.sh and below).
          UNIX_M=$(jq -cn --argjson a "$UNIX" '{include:$a}')
          WIN_M=$(jq -cn --argjson a "$WIN" '{include:$a}')
          PORTABLE_M=$(jq -cn --argjson a "$PORTABLE" '{include:$a}')
          echo "unix=$UNIX_M" >> "$GITHUB_OUTPUT"
          echo "windows=$WIN_M" >> "$GITHUB_OUTPUT"
          echo "portable=$PORTABLE_M" >> "$GITHUB_OUTPUT"

  smoke-unix:
    needs: setup-matrix
    strategy:
      fail-fast: false
      matrix: ${{ fromJSON(needs.setup-matrix.outputs.unix) }}
    runs-on: ${{ matrix.os }}
    # 30, not 15: the unified smoke legitimately runs MORE (artifact mode,
    # real download/update phases, a Phase 15 that actually executes), and
    # v4's PASSING legs already took 12-14 min — v5 lost 13 green-path jobs
    # to the 15-min wall (GitHub reports job timeouts as 'cancelled').
    # Budget doctrine: the timeout sits above the worst case, twice over.
    timeout-minutes: 30
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
        with:
          name: binaries-${{ matrix.goos }}-${{ matrix.goarch }}

      - name: Extract release artifact
        run: |
          mkdir -p "$RUNNER_TEMP/cbm-artifact"
          tar -xzf codebase-memory-mcp-${{ matrix.goos }}-${{ matrix.goarch }}.tar.gz \
            -C "$RUNNER_TEMP/cbm-artifact"
          chmod +x "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"

      # ONE harness for every venue. This is the same wrapper local CI and PR CI
      # run, handed the DOWNLOADED artifact instead of a local build: it stages
      # the release fixture, serves it on a kernel-assigned port, and runs
      # smoke-test.sh under the isolated profile/TEMP the other venues use.
      # It replaces an inline copy of that staging logic which had drifted from
      # the wrapper (no env isolation, no user-PATH guard, fixed port).
      - name: Smoke test (shared wrapper)
        run: |
          scripts/smoke-local.sh "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"
        env:
          CBM_SMOKE_ARTIFACT_DIR: ${{ runner.temp }}/cbm-artifact
          SMOKE_REQUIRE_UI: "1"

      - name: Security audits
        run: |
          BIN="$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"
          scripts/security-strings.sh "$BIN"
          scripts/security-install.sh "$BIN"
          scripts/security-network.sh "$BIN"

      - name: MCP robustness test (selected linux-amd64 runtime)
        if: matrix.goos == 'linux' && matrix.goarch == 'amd64'
        run: |
          BIN="$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"
          scripts/security-fuzz.sh "$BIN"
          scripts/security-fuzz-random.sh "$BIN" 60

      - name: ClamAV scan (Linux)
        if: startsWith(matrix.os, 'ubuntu')
        run: |
          sudo apt-get update -qq && sudo apt-get install -y -qq clamav > /dev/null 2>&1
          # apt auto-starts the clamav-freshclam daemon, which holds a lock on
          # freshclam's log/db; stop it so the manual freshclam below can run
          # (else: "Failed to lock the log file ... Resource temporarily unavailable").
          sudo systemctl stop clamav-freshclam 2>/dev/null || true
          sudo sed -i 's/^Example/#Example/' /etc/clamav/freshclam.conf 2>/dev/null || true
          grep -q "DatabaseMirror" /etc/clamav/freshclam.conf 2>/dev/null || \
            echo "DatabaseMirror database.clamav.net" | sudo tee -a /etc/clamav/freshclam.conf > /dev/null
          sudo freshclam --quiet
          clamscan --no-summary "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"

      - name: ClamAV scan (macOS)
        if: startsWith(matrix.os, 'macos')
        run: |
          brew install clamav > /dev/null 2>&1
          CLAMAV_ETC=$(brew --prefix)/etc/clamav
          if [ ! -f "$CLAMAV_ETC/freshclam.conf" ]; then
            cp "$CLAMAV_ETC/freshclam.conf.sample" "$CLAMAV_ETC/freshclam.conf" 2>/dev/null || true
            sed -i '' 's/^Example/#Example/' "$CLAMAV_ETC/freshclam.conf" 2>/dev/null || true
            echo "DatabaseMirror database.clamav.net" >> "$CLAMAV_ETC/freshclam.conf"
          fi
          freshclam --quiet --no-warnings 2>/dev/null || freshclam --quiet 2>/dev/null || echo "WARNING: freshclam update failed"
          clamscan --no-summary "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"

  smoke-windows:
    needs: setup-matrix
    strategy:
      fail-fast: false
      matrix: ${{ fromJSON(needs.setup-matrix.outputs.windows) }}
    runs-on: ${{ matrix.os }}
    # 30, not 15: the unified smoke legitimately runs MORE (artifact mode,
    # real download/update phases, a Phase 15 that actually executes), and
    # v4's PASSING legs already took 12-14 min — v5 lost 13 green-path jobs
    # to the 15-min wall (GitHub reports job timeouts as 'cancelled').
    # Budget doctrine: the timeout sits above the worst case, twice over.
    timeout-minutes: 30
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

      - uses: msys2/setup-msys2@66cd2cce69caa17b53920067426061ca1de3a884 # v2
        with:
          msystem: ${{ matrix.arch == 'arm64' && 'CLANGARM64' || 'CLANG64' }}
          path-type: inherit
          install: >-
            mingw-w64-clang-${{ matrix.arch == 'arm64' && 'aarch64' || 'x86_64' }}-python3
            mingw-w64-clang-${{ matrix.arch == 'arm64' && 'aarch64' || 'x86_64' }}-curl
            unzip
            zip
            coreutils

      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
        with:
          name: binaries-windows-${{ matrix.arch }}

      - name: Extract release artifact
        shell: msys2 {0}
        run: |
          ARCH=${{ matrix.arch }}
          ARTIFACT_DIR="$(cygpath -u "$RUNNER_TEMP")/cbm-artifact"
          mkdir -p "$ARTIFACT_DIR"
          unzip -o "codebase-memory-mcp-windows-${ARCH}.zip" -d "$ARTIFACT_DIR"
          test -s "$ARTIFACT_DIR/codebase-memory-mcp.exe"
          # ONE executable per runtime set: a payload sibling would mean the
          # removed launcher split came back.
          test ! -e "$ARTIFACT_DIR/codebase-memory-mcp.payload.exe"
          # Pre-flight the image from a profile-rooted directory: the shared
          # runner workspace is deliberately not a valid install root.
          PROFILE_ROOT="$(cygpath -u "$USERPROFILE")"
          LAUNCH_DIR="$(mktemp -d "$PROFILE_ROOT/cbm-release-version.XXXXXX")"
          # Best-effort cleanup with one retry: Windows holds the just-exited
          # image's file busy for ~100ms (section teardown / first-touch AV),
          # and a temp dir on an ephemeral runner must never fail the job.
          trap 'rm -rf "$LAUNCH_DIR" 2>/dev/null || { sleep 2; rm -rf "$LAUNCH_DIR" 2>/dev/null || true; }' EXIT
          cp "$ARTIFACT_DIR/codebase-memory-mcp.exe" "$LAUNCH_DIR/"
          "$LAUNCH_DIR/codebase-memory-mcp.exe" --version

      # ONE harness for every venue: the same script win.sh smoke-install and
      # pr.yml run, handed the DOWNLOADED artifact instead of a local build. It
      # stages the release fixture under a disposable profile root, serves it on
      # a kernel-assigned port, guards AND verifies the user-PATH registry
      # mutation, and neutralizes every agent-config destination before running
      # smoke-test.sh. It replaces an inline copy of that staging which had none
      # of the isolation and never verified the PATH mutation it performed.
      - name: Smoke test (shared wrapper)
        shell: msys2 {0}
        env:
          SMOKE_ARCH: ${{ matrix.arch }}
          SMOKE_REQUIRE_UI: "1"
        run: |
          export CBM_SMOKE_ARTIFACT_DIR="$(cygpath -u "$RUNNER_TEMP")/cbm-artifact"
          bash test-infrastructure/vm/vm-smoke.sh

      - name: Security audits
        shell: msys2 {0}
        run: |
          ARTIFACT_DIR="$(cygpath -u "$RUNNER_TEMP")/cbm-artifact"
          scripts/security-strings.sh "$ARTIFACT_DIR/codebase-memory-mcp.exe"
          PROFILE_ROOT="$(cygpath -u "$USERPROFILE")"
          SECURITY_DIR="$(mktemp -d "$PROFILE_ROOT/cbm-release-security.XXXXXX")"
          # Best-effort cleanup with one retry (see the version preflight).
          trap 'rm -rf "$SECURITY_DIR" 2>/dev/null || { sleep 2; rm -rf "$SECURITY_DIR" 2>/dev/null || true; }' EXIT
          cp "$ARTIFACT_DIR/codebase-memory-mcp.exe" "$SECURITY_DIR/"
          TMPDIR="$SECURITY_DIR" \
            scripts/security-install.sh "$SECURITY_DIR/codebase-memory-mcp.exe"

      - name: Windows Defender scan
        shell: pwsh
        run: |
          & "C:\Program Files\Windows Defender\MpCmdRun.exe" -SignatureUpdate 2>$null
          $artifactDir = Join-Path $env:RUNNER_TEMP "cbm-artifact"
          foreach ($binary in @("codebase-memory-mcp.exe")) {
            $result = & "C:\Program Files\Windows Defender\MpCmdRun.exe" -Scan -ScanType 3 -File (Join-Path $artifactDir $binary) -DisableRemediation
            $code = $LASTEXITCODE
            Write-Host $result
            # MpCmdRun -Scan exit codes: 0 = clean, 2 = threat found. Any OTHER non-zero
            # means the scan engine could not run at all (e.g. hr=0x800106ba: the Defender
            # antimalware service is unavailable on the runner) — that is NOT a detection.
            # Fail soft on an engine failure so a transient runner-side AV outage can't
            # false-fail a release; only a real detection (exit 2) hard-blocks.
            if ($code -eq 2) {
              Write-Host "BLOCKED: Windows Defender flagged $binary!"; exit 1
            } elseif ($code -ne 0) {
              Write-Host "::warning::Windows Defender scan could not run for $binary (exit $code) - skipping AV gate on this runner"
            } else {
              Write-Host "=== Windows Defender: $binary clean ==="
            }
          }

  smoke-linux-portable:
    needs: setup-matrix
    strategy:
      fail-fast: false
      matrix: ${{ fromJSON(needs.setup-matrix.outputs.portable) }}
    runs-on: ${{ matrix.runner }}
    # 30, not 15: the unified smoke legitimately runs MORE (artifact mode,
    # real download/update phases, a Phase 15 that actually executes), and
    # v4's PASSING legs already took 12-14 min — v5 lost 13 green-path jobs
    # to the 15-min wall (GitHub reports job timeouts as 'cancelled').
    # Budget doctrine: the timeout sits above the worst case, twice over.
    timeout-minutes: 30
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
        with:
          name: binaries-linux-${{ matrix.arch }}-portable

      - name: Extract release artifact
        run: |
          mkdir -p "$RUNNER_TEMP/cbm-artifact"
          tar -xzf codebase-memory-mcp-linux-${{ matrix.arch }}-portable.tar.gz \
            -C "$RUNNER_TEMP/cbm-artifact"
          chmod +x "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"

      # Same wrapper as every other venue. This leg previously called
      # smoke-test.sh with no fixture server, which silently SKIPPED the
      # download/checksum/install-script phases (12-13) — the portable binary is
      # what every linux install and update actually delivers, so those are
      # exactly the phases it most needs to run.
      - name: Smoke test (shared wrapper)
        run: |
          scripts/smoke-local.sh "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"
        env:
          CBM_SMOKE_ARTIFACT_DIR: ${{ runner.temp }}/cbm-artifact
          SMOKE_REQUIRE_UI: "1"

      # The -portable binary is what all linux install/update paths now deliver;
      # it MUST start on old glibc (Debian 11 / RHEL 8 / Ubuntu 20.04). Runs it
      # in debian:bullseye (glibc 2.31) — the standard dynamic binary would fail
      # here with `GLIBC_2.38 not found`.
      - name: Old-glibc compatibility
        run: scripts/ci/check-glibc-compat.sh "$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"

      - name: Security audits
        run: |
          BIN="$RUNNER_TEMP/cbm-artifact/codebase-memory-mcp"
          scripts/security-strings.sh "$BIN"
          scripts/security-install.sh "$BIN"
          scripts/security-network.sh "$BIN"

  # ── Packaging check (non-gating) ──────────────────────────────────
  # Builds pkg/glama/Dockerfile — the image Glama builds to score the
  # server — and runs an MCP initialize + tools/list handshake to confirm the
  # containerized stdio server still starts and introspects. Guards the Glama
  # listing integration against drift (Dockerfile breakage, release-asset
  # renames, introspection regressions).
  #
  # Also tests the brew tap/install flow.
  #
  # continue-on-error: a broken *directory* image must never block a release —
  # the shipped product binaries are unaffected. The red X is the signal to fix
  # the integration, not a release gate.
  smoke-packages:
    runs-on: ubuntu-latest
    timeout-minutes: 10
    continue-on-error: true
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
      - name: Build Glama image and verify MCP introspection
        run: bash pkg/glama/verify.sh
      - name: Test homebrew installation
        env:
          HOMEBREW_NO_AUTO_UPDATE: 1
        run: |
          eval "$(/home/linuxbrew/.linuxbrew/bin/brew shellenv)"
          brew tap deusdata/codebase-memory-mcp "$GITHUB_WORKSPACE"
          brew trust deusdata/codebase-memory-mcp
          brew install codebase-memory-mcp
          codebase-memory-mcp --version