From c6abe91fa0097f0330420c4ff08b204f0478888a Mon Sep 17 00:00:00 2001 From: Rue <343175018+thatsrue@users.noreply.github.com> Date: Thu, 19 Nov 2026 07:28:56 -0500 Subject: [PATCH 1/2] docs: the act rule runs on the latest image again, and says what act can't show --- docs/repo-standards.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/repo-standards.md b/docs/repo-standards.md index c227d8b..30e827a 100644 --- a/docs/repo-standards.md +++ b/docs/repo-standards.md @@ -89,7 +89,7 @@ Use only GitHub's default labels and labels added by Dependabot. - Every repo with code has one workflow, `.github/workflows/ci.yml`, named `CI`. Its jobs are `lint`, `typecheck` (typed code), `test` and `build` (anything that ships), plus `e2e` where there are end-to-end tests. A repo whose checks are one script runs it as one job named `test`. - A job never sets a `name:` that differs from its id. Required checks and other workflows key on these names. - Jobs call the repo's own scripts instead of repeating tool commands and flags. -- A change to `.github/workflows/` runs each job it changes with [act](https://github.com/nektos/act) on the local Podman before the push, so a step that fails in CI's container fails here first. Point act at Podman's socket (`DOCKER_HOST=unix://$(podman machine inspect --format '{{.ConnectionInfo.PodmanSocket.Path}}')`) and run `act -j --container-architecture linux/arm64 -P ubuntu-latest=catthehacker/ubuntu:act-latest --container-daemon-socket -`. act can't run JavaScript actions inside a job with `container:` ([nektos/act#732](https://github.com/nektos/act/issues/732)), so a step there that only speeds CI up, like `actions/cache`, sets `if: ${{ !env.ACT }}`. A job act can't run, like a macOS job, is named in the pull request with how it was checked instead. +- A change to `.github/workflows/` runs each job it changes with [act](https://github.com/nektos/act) on the local Podman before the push, so a step that fails in CI's container fails here first. Run it from a plain clone: act can't read a linked worktree's `.git` ([nektos/act#6074](https://github.com/nektos/act/issues/6074)). Point act at Podman's socket (`DOCKER_HOST=unix://$(podman machine inspect --format '{{.ConnectionInfo.PodmanSocket.Path}}')`) and run `act -j --container-architecture linux/arm64 -P ubuntu-latest=catthehacker/ubuntu:act-latest --container-options '--security-opt label=disable' --container-daemon-socket -`. The label option is needed because act keeps its container up with `tail -f /dev/null`, which Ubuntu 26.04's `tail` can't do under the Podman machine's SELinux labels. act can't run JavaScript actions inside a job with `container:` ([nektos/act#732](https://github.com/nektos/act/issues/732)), so a step there that only speeds CI up, like `actions/cache`, sets `if: ${{ !env.ACT }}`. A job act can't run, like a macOS job, is named in the pull request with how it was checked instead. act's image holds more tools than a self-hosted runner's ([`actions-runner`](https://docs.github.com/en/actions/concepts/runners/actions-runner-controller) is minimal by design), so a green act run doesn't show that a tool exists there: a step installs what it calls or uses a setup action. A test that fails only on Linux isn't caught before the push either; CI is the gate for that. - Run on pull requests and on pushes to `main`. Add `workflow_dispatch` when people or other workflows start it by hand, and never remove it without checking what starts it. A job-level `if:` must allow every trigger the workflow has. - Set `permissions: contents: read` at the top. A job that needs more asks for it itself. - Every job sets `timeout-minutes`, based on how long it really takes, with room to spare. Jobs that call a reusable workflow are exempt. From d6d91fd876b78476f88fc59504497af17d7f300d Mon Sep 17 00:00:00 2001 From: Rue <343175018+thatsrue@users.noreply.github.com> Date: Thu, 19 Nov 2026 07:31:53 -0500 Subject: [PATCH 2/2] docs: split the act rule into one bullet per point and name what the label option turns off --- docs/repo-standards.md | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/docs/repo-standards.md b/docs/repo-standards.md index 30e827a..bdae4e7 100644 --- a/docs/repo-standards.md +++ b/docs/repo-standards.md @@ -89,7 +89,12 @@ Use only GitHub's default labels and labels added by Dependabot. - Every repo with code has one workflow, `.github/workflows/ci.yml`, named `CI`. Its jobs are `lint`, `typecheck` (typed code), `test` and `build` (anything that ships), plus `e2e` where there are end-to-end tests. A repo whose checks are one script runs it as one job named `test`. - A job never sets a `name:` that differs from its id. Required checks and other workflows key on these names. - Jobs call the repo's own scripts instead of repeating tool commands and flags. -- A change to `.github/workflows/` runs each job it changes with [act](https://github.com/nektos/act) on the local Podman before the push, so a step that fails in CI's container fails here first. Run it from a plain clone: act can't read a linked worktree's `.git` ([nektos/act#6074](https://github.com/nektos/act/issues/6074)). Point act at Podman's socket (`DOCKER_HOST=unix://$(podman machine inspect --format '{{.ConnectionInfo.PodmanSocket.Path}}')`) and run `act -j --container-architecture linux/arm64 -P ubuntu-latest=catthehacker/ubuntu:act-latest --container-options '--security-opt label=disable' --container-daemon-socket -`. The label option is needed because act keeps its container up with `tail -f /dev/null`, which Ubuntu 26.04's `tail` can't do under the Podman machine's SELinux labels. act can't run JavaScript actions inside a job with `container:` ([nektos/act#732](https://github.com/nektos/act/issues/732)), so a step there that only speeds CI up, like `actions/cache`, sets `if: ${{ !env.ACT }}`. A job act can't run, like a macOS job, is named in the pull request with how it was checked instead. act's image holds more tools than a self-hosted runner's ([`actions-runner`](https://docs.github.com/en/actions/concepts/runners/actions-runner-controller) is minimal by design), so a green act run doesn't show that a tool exists there: a step installs what it calls or uses a setup action. A test that fails only on Linux isn't caught before the push either; CI is the gate for that. +- A change to `.github/workflows/` runs each job it changes with [act](https://github.com/nektos/act) on the local Podman before the push, so a step that fails in CI's container fails here first. Point act at Podman's socket (`DOCKER_HOST=unix://$(podman machine inspect --format '{{.ConnectionInfo.PodmanSocket.Path}}')`) and run `act -j --container-architecture linux/arm64 -P ubuntu-latest=catthehacker/ubuntu:act-latest --container-options '--security-opt label=disable' --container-daemon-socket -`. +- Without `--security-opt label=disable`, act's keep-alive command (`tail -f /dev/null`) is denied under the Podman machine's SELinux labels on Ubuntu 26.04 and the job container exits at start. The option turns off SELinux labels for that throwaway container only. +- Run act from a plain clone. In a linked worktree `.git` is a file that points at a host path, which the container can't follow ([nektos/act#6074](https://github.com/nektos/act/issues/6074)). +- act can't run JavaScript actions inside a job with `container:` ([nektos/act#732](https://github.com/nektos/act/issues/732)), so a step there that only speeds CI up, like `actions/cache`, sets `if: ${{ !env.ACT }}`. A job act can't run, like a macOS job, is named in the pull request with how it was checked instead. +- A green act run doesn't show that a tool exists on the runner. act's image has more tools than the [ARC runner image](https://docs.github.com/en/actions/concepts/runners/actions-runner-controller), which is minimal by design, so a step installs what it calls or uses a setup action. +- A test that passes on macOS and fails only on Linux isn't caught by this check. CI is the gate for it. - Run on pull requests and on pushes to `main`. Add `workflow_dispatch` when people or other workflows start it by hand, and never remove it without checking what starts it. A job-level `if:` must allow every trigger the workflow has. - Set `permissions: contents: read` at the top. A job that needs more asks for it itself. - Every job sets `timeout-minutes`, based on how long it really takes, with room to spare. Jobs that call a reusable workflow are exempt.