Skip to content

A workflow red on the default branch for many runs goes unseen when it is not a required check #387

Closed

Description

@napkinstack-agent

Who asks, and what they cannot do today

The decider, or the session closing a cycle, does not learn that a workflow on the default branch keeps failing. On GitHub, a workflow outside the ruleset's required checks can be red on every run on main while every pull request merges green. Neither nstack plan nor nstack doctor says so.

Evidence

On NapkinStack/kitokey, the Windows packaging workflow failed on every run on main from 2026-11-07 (a link error, LNK1120). It was noticed on 2026-11-13, six runs later, by a verifier looking at something else. It is not a required check, so every merge stayed CLEAN. Its paths: filter also leaves out a module its jobs build, so a pull request touching that module never ran it.

What happens if nothing is done

A platform the product ships on stays broken, unseen, for a week or more, until a release. The later it is seen, the more merges sit on top of the break. On kitokey the fix also needed the person, since agents cannot dispatch workflows.

Smallest answers for the framing to weigh:

  • nstack doctor (or plan) reads the default branch's last N workflow runs, through the forge API with the token it already has, and warns when one workflow's last N conclusions are all failures, naming it and its first failing run;
  • later, and only if it pays: a fitness check warning when a workflow's paths: filter leaves out a module its jobs build.

Activity

  1. napkinstack-agent commented on Nov 14, 2026

    @napkinstack-agent
    ContributorAuthor

    Triage: generic. Any repository whose non-required workflows run on the default branch. Market: GitHub shows a red run but notifies nobody once it is not required; Dependabot and Renovate dashboards have the same blind spot. Slot: after D216, as D217. Direction: nstack doctor reads the default branch's last runs per workflow through the API and token it already has, and warns when the last N all failed, naming the first failing run. The paths: filter check is set aside until it pays.

  2. napkinstack-agent commented on Nov 14, 2026

    @napkinstack-agent
    ContributorAuthor

    Slot moved: D218 (D217 went to a test-isolation fix, #393 if numbered so). The direction is unchanged.

  3. napkinstack-agent commented on Nov 14, 2026

    @napkinstack-agent
    ContributorAuthor

    Slot moved again: D219. #394 blocks merges and goes first as D218.

  4. napkinstack-agent commented on Nov 14, 2026

    @napkinstack-agent
    ContributorAuthor

    D219 merged in 31a170c (#396). Unreleased, it goes in v0.29.0.

  5. napkinstack-agent commented on Nov 14, 2026

    @napkinstack-agent
    ContributorAuthor

    Confirmed on live data: D219's _default_branch_runs was fed NapkinStack/kitokey's real workflow runs, read through the forge API. It says:

    RED            Windows app: failed its last 6 or more runs on main, since 2026-11-13 — information, not a gap
                   First failure: https://github.com/NapkinStack/kitokey/actions/runs/43492520873
    

    The cancelled run in the streak is neither a failure nor a pass, as designed. The end-to-end confirmation, nstack doctor run with a token, remains for the release.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions