hermes-agent/.github/workflows/pm-bundle.yml

name: PM Bundle

# Stages the self-contained agent payload for each (os, arch) target with
# `hermes pm bundle`: repo snapshot + tool store + facts + a relocatable
# venv built on the staged python-build-standalone interpreter. Each
# payload is smoke-tested on its native runner — the staged interpreter
# boots hermes_cli out of the payload with no network and no PYTHONPATH —
# as a pre-merge check for the pm tooling itself.
#
# Staging is native per target (no cross-target wheel tables anywhere):
# the runner IS the target. pm's arch guard fails the job if any staged
# mismatches, before anything ships.

on:
  workflow_dispatch:
    inputs:
      ref:
        description: 'Git ref to bundle (default: the triggering commit, github.sha)'
        required: false
        type: string
  workflow_call:
    inputs:
      release:
        description: 'Stable-release candidate run (ignored by the jobs; uniform callable surface).'
        required: false
        type: boolean
        default: false

      ref:
        description: 'Git ref to bundle (default: the triggering commit, github.sha)'
        required: false
        type: string
        default: ''
  pull_request:
    paths:
      - 'pm/**'
      - 'scripts/bundles/**'
      - 'scripts/ci/desktop_build_cache.py'
      - '.github/actions/desktop-build-cache/**'
      - 'scripts/windows-build-deps.ps1'
      - 'scripts/build/windows-deps.ps1'
      - '.github/actions/setup-windows-build-deps/**'
      - 'uv.lock'
      - 'pyproject.toml'
      - '.github/workflows/pm-bundle.yml'

permissions:
  contents: read

# A workflow_call run (stable release) is never cancelled: its group uses
# github.run_id so a parent rerun cannot kill this child mid-flight.
concurrency:
  group: pm-bundle-${{ inputs.release == true && github.run_id || github.ref }}
  cancel-in-progress: ${{ inputs.release != true }}

jobs:
  bundle:
    name: bundle ${{ matrix.target.label }}
    runs-on: ${{ matrix.target.runner }}
    # Leave time for setup, cache saves and smoke tests around the wheel build.
    timeout-minutes: 180
    strategy:
      fail-fast: false
      matrix:
        target:
          # TODO(ethie): restore the linux legs once the ubuntu-latest-32-*
          # larger runners are re-imaged to Ubuntu 24.04. They currently boot
          # ubuntu-22.04 (glibc 2.35), and the prebuilt llama-server staged by
          # the native payload requires GLIBC_2.38, so verification fails.
          # - label: linux-x64
          #   runner: ubuntu-latest-32-core
          # - label: linux-arm64
          #   runner: ubuntu-latest-32-arm-core
          - label: darwin-arm64
            runner: macos-15
          - label: darwin-x64
            runner: macos-15-intel
          - label: win32-x64
            runner: windows-latest-32-core
          - label: win32-arm64
            runner: windows-latest-32-arm-core
    steps:
      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          # Default to the exact candidate commit (github.sha); inputs.ref
          # remains an escape hatch for a focused manual reproduction.
          ref: ${{ inputs.ref || github.sha }}
          fetch-tags: true

      # The bootstrap scripts need Python >= 3.11 (tomllib). Larger runners
      # (ubuntu-latest-32-*) are provisioned with ubuntu-22.04 images whose
      # system python3 is 3.10, so pin the bootstrap interpreter instead of
      # trusting the image.
      - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
        with:
          python-version: "3.13"

      - name: Restore native dependency candidates
        id: native-cache
        uses: ./.github/actions/desktop-build-cache
        with:
          phase: restore
          source: ${{ github.workspace }}
          work: ${{ runner.temp }}/payload-job
          cache: ${{ runner.temp }}/payload-inputs
          producer: payload-test

      - name: Prepare native payload dependencies
        id: prepare
        shell: bash
        env:
          PAYLOAD_WORK: ${{ runner.temp }}/payload-job
          PAYLOAD_CACHE: ${{ runner.temp }}/payload-inputs
          PYTHONUTF8: '1'
        run: |
          if [ "$RUNNER_OS" = Windows ]; then bootstrap=python; else bootstrap=python3; fi
          "$bootstrap" -S -B scripts/bundles/native_build.py --prepare-only \
            --source "$GITHUB_WORKSPACE" --work "$PAYLOAD_WORK" --cache "$PAYLOAD_CACHE" \
            --out "$GITHUB_WORKSPACE/build/agent-payload" --ref HEAD

      # PRs can restore candidates but never gain a cache-writing lane here.
      - name: Save native dependency candidates before assembly
        if: ${{ !cancelled() && steps.prepare.outcome == 'success' && github.event_name != 'pull_request' && github.ref == 'refs/heads/main' && (inputs.ref == '' || inputs.ref == github.sha) }}
        uses: ./.github/actions/desktop-build-cache
        with:
          phase: save
          source: ${{ github.workspace }}
          work: ${{ runner.temp }}/payload-job
          cache: ${{ runner.temp }}/payload-inputs
          producer: payload-test
          key: ${{ steps.native-cache.outputs.cache-key }}

      - name: Assemble the prepared payload without dependency acquisition
        shell: bash
        run: |
          if [ "$RUNNER_OS" = Windows ]; then bootstrap=python; else bootstrap=python3; fi
          "$bootstrap" -S -B scripts/bundles/native_build.py \
            --prepared "$GITHUB_WORKSPACE/build/agent-payload.prepared.json"

      # The generated command must survive relocation without checkout env/cwd.
      - name: Smoke test the relocated payload (network disabled)
        if: startsWith(matrix.target.label, 'linux-')
        shell: bash
        run: sudo --preserve-env=PATH unshare --net bash scripts/smoke-payload.sh build/agent-payload

      - name: Smoke test the relocated payload (native launchers)
        if: ${{ !startsWith(matrix.target.label, 'linux-') }}
        shell: bash
        run: bash scripts/smoke-payload.sh build/agent-payload