mirror of
https://github.com/EveryInc/compound-engineering-plugin.git
synced 2026-09-19 01:09:55 +08:00
Trevin Chow
193 lines
8.0 KiB
YAML
193 lines
8.0 KiB
YAML
name: CI
|
|
|
|
on:
|
|
push:
|
|
branches: [main]
|
|
pull_request:
|
|
workflow_dispatch:
|
|
|
|
jobs:
|
|
pr-title:
|
|
if: github.event_name == 'pull_request'
|
|
runs-on: ubuntu-latest
|
|
permissions:
|
|
pull-requests: read
|
|
|
|
steps:
|
|
- name: Validate PR title
|
|
uses: amannn/action-semantic-pull-request@v6.1.1
|
|
env:
|
|
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
with:
|
|
requireScope: false
|
|
types: |
|
|
feat
|
|
fix
|
|
docs
|
|
refactor
|
|
chore
|
|
test
|
|
ci
|
|
build
|
|
perf
|
|
revert
|
|
|
|
test:
|
|
runs-on: ubuntu-latest
|
|
# A wedged bun --parallel worker (oven-sh/bun#34069) can sit with no output
|
|
# forever; green runs take about 4 minutes and a timeout-stack run with the
|
|
# serial re-run about 15. Fail well before GitHub's 6-hour default so the
|
|
# babysit loop can see it and rerun.
|
|
timeout-minutes: 30
|
|
|
|
env:
|
|
# Single source of truth for the pinned `claude plugin validate` version:
|
|
# both the install and its cache key read this, so a bump cannot bump one
|
|
# and leave the other serving a stale cached binary.
|
|
CLAUDE_CODE_VERSION: 2.1.220
|
|
|
|
steps:
|
|
- uses: actions/checkout@v6
|
|
# Full history so release:validate can read the base-branch (origin/main)
|
|
# release-please manifest when checking release-as pins for staleness.
|
|
with:
|
|
fetch-depth: 0
|
|
|
|
- name: Setup Bun
|
|
uses: oven-sh/setup-bun@v2
|
|
with:
|
|
bun-version: latest
|
|
|
|
- name: Install dependencies
|
|
run: bun install
|
|
|
|
- name: Validate release metadata
|
|
run: bun run release:validate
|
|
|
|
# Canonical schema check: runs `claude plugin validate --strict` against
|
|
# the marketplace catalog and the plugin manifest. The CLI must be
|
|
# installed via npm, not bun: `bun x` cannot run this package (bun blocks
|
|
# its required postinstall and cannot resolve its `claude` bin name), and
|
|
# `bun run` rewrites `npx` in package.json scripts to the broken `bun x`,
|
|
# so the script invokes a `claude` already on PATH instead.
|
|
#
|
|
# The validator version is pinned deliberately for reproducible CI: a
|
|
# floating @latest would track new upstream validation rules
|
|
# automatically but could break CI without any repo change. Bump the
|
|
# pin intentionally to adopt new `claude plugin validate` rules.
|
|
#
|
|
# plugin:validate must hit .claude-plugin/plugin.json — `validate .` only
|
|
# checks the marketplace. Root CLAUDE.md is a symlink to AGENTS.md so
|
|
# --strict stays green; see AGENTS.md "CI and Quality Gates".
|
|
#
|
|
# The install is cached because the npm package is a 165KB wrapper whose
|
|
# postinstall pulls a platform binary (`@anthropic-ai/claude-code-linux-x64`,
|
|
# ~275MB unpacked). Registry throughput for that payload is the whole cost
|
|
# of this step — validation itself is under a second. Observed range on
|
|
# ubuntu-latest: 3-10s typical, 5m worst case. Caching buys predictability,
|
|
# not average speed: it caps the tail rather than beating the fast path.
|
|
# Keying on the pinned version means a hit on every run between bumps.
|
|
- name: Cache claude-code validator
|
|
id: cache-claude-code
|
|
uses: actions/cache@v6
|
|
with:
|
|
path: ~/.npm-global
|
|
key: claude-code-${{ runner.os }}-${{ env.CLAUDE_CODE_VERSION }}
|
|
|
|
- name: Install claude-code validator
|
|
if: steps.cache-claude-code.outputs.cache-hit != 'true'
|
|
run: npm install -g --prefix ~/.npm-global @anthropic-ai/claude-code@${{ env.CLAUDE_CODE_VERSION }}
|
|
|
|
- name: Validate plugin schema (claude plugin validate)
|
|
run: |
|
|
export PATH="$HOME/.npm-global/bin:$PATH"
|
|
claude --version
|
|
bun run plugin:validate
|
|
|
|
# Runs the package `test` script so the flags CI uses and the flags a
|
|
# contributor gets from `bun run test` cannot drift apart.
|
|
- name: Run tests
|
|
run: bun run test
|
|
|
|
# Focused Windows gate for the skill scripts that reach for POSIX process and
|
|
# file primitives: peer-job-runner (#1243) and ce-babysit-pr's pr-snapshot
|
|
# (#1280). Not a full-suite matrix — POSIX lifecycle tests (fork/killpg/ps)
|
|
# stay on ubuntu-latest. This job runs the cross-platform fixtures under real
|
|
# win32, which is the only way to execute the msvcrt/ctypes branches at all:
|
|
# they are imported solely under sys.platform == "win32", so no amount of
|
|
# patching reaches them from Linux or Mac.
|
|
windows-native:
|
|
runs-on: windows-latest
|
|
# Green runs finish in about 90 seconds. The Windows smoke step below buffers
|
|
# its output to a file and can hang with nothing printed when a detached
|
|
# supervisor never exits (2026-09-14, PR #1711 run 34909130752: 25+ minutes
|
|
# stuck, unresponsive to cancel and force-cancel). Fail well before GitHub's
|
|
# 6-hour default so the babysit loop can see it and rerun; 10 minutes still
|
|
# covers the smoke step's three-attempt ctypes retry.
|
|
timeout-minutes: 10
|
|
|
|
steps:
|
|
# Keep LF in the working tree so newline-anchored parity regexes match.
|
|
# Default Windows runners often have core.autocrlf=true (see
|
|
# docs/solutions/developer-experience/windows-crlf-checkout-breaks-newline-anchored-tests.md).
|
|
- name: Disable autocrlf for LF checkout
|
|
run: git config --global core.autocrlf false
|
|
|
|
- uses: actions/checkout@v6
|
|
|
|
- name: Setup Python
|
|
uses: actions/setup-python@v6
|
|
with:
|
|
# Match the verified Windows smoke environment for #1243.
|
|
python-version: "3.11"
|
|
|
|
- name: Setup Bun
|
|
uses: oven-sh/setup-bun@v2
|
|
with:
|
|
bun-version: latest
|
|
|
|
- name: Install dependencies
|
|
run: bun install
|
|
|
|
- name: Peer-job-runner unit fixture
|
|
# Use `python`, never `python3` — on Windows the latter is often the
|
|
# Microsoft Store stub (see docs/solutions/conventions/resolve-python-interpreter-not-python3.md).
|
|
run: python tests/fixtures/peer-job-runner-unit.py
|
|
|
|
- name: Peer-job-runner Windows smoke
|
|
# Hosted windows-latest Python intermittently fails `import ctypes`
|
|
# mid-suite (DLL init on _ctypes) after earlier tests already passed.
|
|
# Product code cannot recover. Retries are signature-gated: only
|
|
# `DLL load failed while importing _ctypes` is retried; any other
|
|
# nonzero exit fails this step immediately. A retried attempt's log is
|
|
# suppressed so a green step never shows failure dumps; the full log
|
|
# prints on success and on the terminal failure.
|
|
shell: pwsh
|
|
run: |
|
|
$max = 3
|
|
for ($i = 1; $i -le $max; $i++) {
|
|
$log = Join-Path $env:TEMP "peer-job-runner-windows-smoke-$i.log"
|
|
python tests/fixtures/peer-job-runner-windows-smoke.py *> $log
|
|
$code = $LASTEXITCODE
|
|
$text = Get-Content -Raw -ErrorAction SilentlyContinue $log
|
|
if (($code -eq 0) -or ($text -notmatch 'DLL load failed while importing _ctypes') -or ($i -eq $max)) {
|
|
if ($null -ne $text) { Write-Host $text }
|
|
exit $code
|
|
}
|
|
Write-Host "Windows smoke hit known _ctypes flake (attempt $i/$max, exit $code); output suppressed, retrying"
|
|
Start-Sleep -Seconds 15
|
|
}
|
|
|
|
- name: Peer-job-runner parity
|
|
run: bun test tests/peer-job-runner-parity.test.ts
|
|
|
|
- name: pr-snapshot platform primitives
|
|
run: python tests/fixtures/pr-snapshot-platform.py
|
|
|
|
# Runs the scratch-root preamble exactly as the skills ship it, under Git Bash. The
|
|
# POSIX suite cannot cover this: `install -d -m 700` works there and fails only here
|
|
# (#1285), so a shape assertion stayed green while every skill's scratch setup aborted
|
|
# on Windows. Shell out via bun so the same file also runs in the ubuntu suite.
|
|
- name: Scratch-root preamble executes under Git Bash
|
|
run: bun test tests/scratch-root-preamble-executes.test.ts
|