Choose proportional OpenClaw tests and checks, diagnose failures, and route environment-sensitive or release proof to its owner.
复制下面这句话,粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具,它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-openclaw-testing-b2c8677a46d2 ,按照其中的说明把「openclaw-testing」安装到你(当前 AI 工具)中。执行前先告诉我将运行的命令和写入的位置,等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件,由原作者维护。在 GitHub 查看
Prove the changed contract with the smallest meaningful check, complete required
checks, then finish. Broaden or repeat only for new changes, failures, or
unresolved risks. Do not add tests that merely mirror reversible, low-impact
implementation changes; use $test-audit when authoring or reviewing tests.
For ordinary local tests, start at docs/reference/test.md#routine-local-order
and #core-commands; read docs/ci.md when CI scope or runner behavior matters. Follow the touched subtree's AGENTS.md.
Trusted development tests, changed checks, typecheck/lint, and builds run locally, including broader suites when the contract warrants them. Remote compute is for isolation, clean installation, packaging, Docker, live services, desktop/platform behavior, or an explicit operator request.
| Change or question | Starting point |
|---|---|
| Runtime defect | Reproduce narrowly; rerun that proof after the repair, plus relevant siblings |
| Trusted source diff | pnpm changed:lanes --json, pnpm check:changed, focused tests |
| Public SDK/plugin contract | Changed checks plus representative consumer tests; no automatic all-plugin sweep |
| Build output, lazy imports, package boundaries | Include pnpm build |
| Workflow | git diff --check and pnpm check:workflows |
| Documentation only | Relevant docs/link/format sanity and git diff --check; no runtime tests |
For specialized proof, load only the selected route:
$crabbox, with the
OpenClaw bootstrap binding below.$release-openclaw-ci. A narrow green rerun
does not itself authorize publication. Do not substitute moving main for
the recorded candidate or Tooling SHA.$release-openclaw-plugin-testing.$openclaw-docker-e2e-authoring.$control-ui-e2e; mock-Gateway boundary proof is
valid when it covers the changed path. State concrete live-proof gaps.Untrusted contributor/fork tooling must never execute locally, including its
wrapper or config. Use secretless fork CI or sanitized direct AWS under
$crabbox; never credential-hydrated Testbox. Credentialed execution requires
maintainer approval after review, and never hydrates an untrusted lease.
For untrusted OpenClaw AWS proof, supply the clean trusted main copy of
scripts/crabbox-untrusted-bootstrap.sh as Crabbox's
<trusted-bootstrap-script>. Bind the fresh lease and --fresh-pr checkout to
the reviewed full head SHA. The trusted bootstrap verifies that SHA, the IMDSv2
no-role boundary, and the package-manager pin before installing into an isolated
HOME. Keep CRABBOX_ENV_ALLOW=CI, --no-hydrate, no instance role, and no
Tailscale. A moved head needs a fresh lease; missing no-role proof or no remote
PR means secretless CI. Read the Crabbox untrusted procedure before allocation.
For trusted remote proof, use node scripts/crabbox-wrapper.mjs with the
resolved provider; do not silently switch providers or bypass sync/security
exclusions. Save and reuse task-owned leases, verify the materialized candidate,
keep evidence outside the synced checkout, and stop owned leases at handoff.
Use isolated state and a free port. Never restart, edit, or test against an operator Gateway or real data without explicit per-task approval. Do not kill unrelated processes or reconcile a shared dependency install while jobs use it.
pnpm changed:lanes --json
pnpm check:changed
pnpm test <path-or-filter>
pnpm test:changed
check:changed selects formatting, typecheck, lint, and guard work and may run
targeted owner Vitest tests. Inspect its plan with
node scripts/check-changed.mjs --dry-run -- <paths...>; it is not the full test
suite. test:changed chooses direct tests, mapped/sibling tests, and import
dependents; shared harness/config edits can need explicit targets or the broad
fallback OPENCLAW_TEST_CHANGED_BROAD=1 pnpm test:changed.
pnpm verify runs the full check and then test when that scope is justified.
Use repository wrappers rather than raw Vitest so project routing and setup remain correct. If dependencies are ready in a linked worktree, these bypass pnpm dependency reconciliation:
node scripts/check-changed.mjs
node scripts/run-vitest.mjs <path-or-filter>
Concurrent test/check commands must not share a Vitest filesystem module cache:
serialize them, group tests in one invocation, or give each command a distinct
OPENCLAW_VITEST_FS_MODULE_CACHE_PATH. Checks can schedule Vitest too.
For worker-sensitive failures, OPENCLAW_VITEST_MAX_WORKERS=1 pnpm test <path>
provides a focused serial probe; do not make a forced environment the repair.
gh run list --branch <branch> --limit 10 --json databaseId,headSha,status,conclusion,url
gh run view <run-id> --json status,conclusion,headSha,url,jobs
gh run view <run-id> --job <failed-job-id> --log
Bind the diagnosis to the exact SHA and job. Check whether cancellation means a newer same-branch run superseded it. Fetch relevant failed logs once and reuse them; prefer exact run/job state over a stale PR rollup. Separate product, harness, infrastructure, and credential failures before choosing a retry.
For prompt snapshot drift that passes on macOS, reproduce in CI's Linux/Node environment before regenerating; a local pass cannot override failing CI bytes. Fix related failures and rerun the affected proof. Route unrelated failures with evidence rather than broadening this task automatically.