DEV Community

Cover image for GitHub's December Workflow Change: A Freelancer's Brownout Handoff Plan
Marcus Kim
Marcus Kim

Posted on Originally published at marcusykim.medium.com

GitHub's December Workflow Change: A Freelancer's Brownout Handoff Plan

A client's automation turns red. The app has not changed. An AI assistant offers to rewrite the workflow, and suddenly a small maintenance task becomes a risky afternoon of edits.

Before I would approve that work, I would ask a different question: was the workflow supposed to be allowed to run at that moment?

On November 2, GitHub announced an updated enforcement date for its default pull_request_target protection. For affected public repositories, the default block starts December 2, 2026, unless an applicable Actions event policy allows the event. The previously announced enforcement date was November 2. Private and internal repositories are unaffected; the change covers GitHub Free, Team, and Enterprise Cloud, not Enterprise Server.

GitHub scheduled two temporary enforcement windows for November 9 and November 16, both from 16:00 to 18:00 UTC. During these brownouts, blocked attempts produce failed runs with an error annotation, but the workflows do not execute. Afterward, the default rule returns to evaluate mode until permanent enforcement.

For a freelancer, that is a useful rehearsal opportunity. The deliverable is not simply a green check. It is a plan that lets the next person recognize the expected failure, make the right decision, and avoid undoing a protection in a panic.

This article was prepared with AI assistance. The client scenario and rehearsal below are proposed exercises, not claims about a production migration I performed.

Start with the job, not the YAML file

Imagine a small public app repository with an automation that labels incoming pull requests. A second workflow tests proposed changes. They may both react to a pull request, but they do different jobs and do not automatically need the same privileges.

I would ask the maintainer to name the visible result of each workflow before editing either one. Does it label a request, run tests, publish a preview, or change a release? Who notices when that result is missing?

That small inventory keeps a broad instruction such as "fix CI" from authorizing unrelated changes.

For each relevant workflow, I would record its file path, trigger, intended result, person responsible for the policy decision, and a link to a representative run. Do not paste secrets or private payloads into the handoff note.

A failed run and executed code are different facts

The dangerous shortcut is treating every red result as evidence that application code ran and failed.

GitHub's workflow execution documentation describes Policy insights for examining runs blocked by active policies or identified by evaluate policies. It also explains that protections can be scoped to particular workflows and layered across repository, organization, and enterprise levels. Availability of a custom Evaluate setting depends on the plan; do not assume every account has identical controls.

My handoff would preserve the run's actual annotation and policy explanation. Then I would separate three observations: an attempt was made, the provider permitted or blocked execution, and the intended user-facing result did or did not occur.

That distinction changes the repair. A blocked execution needs a policy and workflow review. A permitted run with a failed test needs investigation of that test. Editing application code before making that distinction creates noise without resolving the cause.

Do not make the exception the default answer

GitHub's security guide for pull_request_target explains why this trigger needs care: it runs with the base repository's trust and can access secrets. Checking out and then executing untrusted pull-request code can expose that privileged context. Test commands, build scripts, dependencies, and configuration can all execute code.

The guide recommends considering a safer event such as pull_request when elevated access is unnecessary, or separating untrusted code handling from privileged work. Changing the event name alone is not a completed migration; its permissions and behavior still need review.

For the hypothetical labeling task, I would ask why access is needed and whether the workflow ever executes contributed code. For the test task, I would ask whether its purpose can be met without privileged execution. Those answers belong in the decision note before someone proposes an exception.

If an exception is genuinely necessary, an authorized maintainer should review its exact scope. "Allow everything so the checks go green" is not a maintenance strategy. A green result obtained by weakening an unrelated protection is a different outcome from the one the client requested.

Give the rehearsal a beginning and an end

Here is the compact handoff I would prepare for a brownout. Both November windows have passed. Use any saved annotations to prepare for the December 2 change:

  1. Scope: identify the affected repository and workflow, plus the user-facing result it owns.
  2. Expectation: describe the policy signal expected during the window and the work that should not execute if blocked.
  3. Decision: record whether the workflow is being redesigned, intentionally left blocked, or reviewed for a narrow exception.
  4. Owner: name the person who can approve that decision and the person checking the result.
  5. Evidence: link the relevant run, annotation, policy view, and reviewed change without exposing secrets.
  6. Follow-up: state what to check after the temporary window ends and before permanent enforcement.

Use the provider's UTC times in the note and add the team's local time explicitly. I would not rely on a calendar entry that silently changes meaning across time zones.

For a real repository, rehearse only within its existing authorization and safe testing practices. Do not create privileged test workflows or weaken security settings merely to demonstrate that a checklist can be completed.

Close the handoff, not just the notification

After a temporary window, a workflow running again does not prove that December's decision has been made. The temporary enforcement may simply have ended.

I would close the task only when the owner can explain the selected long-term behavior and point to the evidence supporting it. If the answer is still unknown, the honest status is unresolved with an owner and next step, not fixed.

That is the reusable freelancer lesson: use a provider's scheduled disruption to practice diagnosis and ownership before it becomes an urgent client problem. Keep the rehearsal small enough that someone else can actually follow it.

Top comments (0)