メニュー
Run this repo's pytest suite safely. Use for ANY pytest invocation here — an uncapped run once grew to 30GB and OOM-killed the machine. Covers the memory cap, orphan cleanup, and the mock-stream failure mode behind it.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/protella/chatgpt-bots --skill run-testsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Drive the dev Slack bot end-to-end against real Slack. Use when verifying bot behavior live — participation/wake decisions, reactions, streaming, image and file paths — or when a unit test passes but the real path is suspect. Covers the user-token harness, the durable live battery, and the contamination traps.
The standing workflow for any substantial change in this repo — codex alignment on the plan, sub-agents for the build, codex review of the diff, then capped tests and a live dev-bot pass. Use before starting any medium-or-larger implementation.
SOC 職業分類に基づく
SKILL.md を表示中
| name | run-tests |
| description | Run this repo's pytest suite safely. Use for ANY pytest invocation here — an uncapped run once grew to 30GB and OOM-killed the machine. Covers the memory cap, orphan cleanup, and the mock-stream failure mode behind it. |
Every pytest invocation gets a virtual-memory cap. No exceptions, including single-test runs.
(ulimit -v 4194304; timeout 600 python3 -m pytest tests/unit -q -p no:cacheprovider --tb=no -p no:logging)
make test (coverage) and make test-fast are fine to reach for, but wrap them the same way.
An uncapped python3 -m pytest tests/unit reached 30 GB RSS and the kernel OOM-killed it,
taking WSL down with it — repeatedly. The cap costs nothing and converts a runaway into a
MemoryError traceback that names the culprit test instead of a dead machine.
--tb=no is not sufficient on its own: assertion rewriting builds the failure explanation
inside the test, before any reporting happens. The ulimit is the real guard.
Interrupting the agent turn does not kill the process it spawned. A stopped turn left an orphaned pytest eating the box for another 400 seconds.
pgrep -af pytest before and after any run.pkill alone — inside a compound command it kills the
calling shell (exit 144).A mock stream that never terminates. output_text += <MagicMock> does not raise — it silently
turns output_text into a mock, and a stale side_effect once left an async iterator with no end.
Any test mocking a response stream must terminate and must yield real strings. More in
Docs/TOOL_SUBSYSTEMS.md.