Restore Actions capacity so the pending fixes can merge #38
Description
What to build
Every private repo in the estate is currently unable to run any workflow. Runs are created, allocated nothing, and fail in ~3 seconds with zero steps executed. Public repos are unaffected. The org exhausted its included Actions minutes on 2026-08-27 at roughly 10:40 UTC.
Authoritative figures for August 2026, month-to-date, from the billing API:
| repo | minutes |
|---|---|
| fleet-infra | 2,235 |
| workspace | 472 |
| nix-config | 217 |
| others | small |
| private total | ~2,975 of 3,000 |
Sixteen pull requests are open that would take this to roughly 800/month. None of them can merge. Every repo's ruleset requires exactly one status check, Pipeline Complete, and that check cannot report while there are no runners. The fix is gated by the problem it fixes.
Break the deadlock so the pending work can land. Two routes, either is fine:
- Raise or remove the Actions spending limit in the organization's billing settings, which restores runners immediately.
- Merge the three highest-value PRs by administrative bypass of the ruleset, then let the restored capacity carry the rest.
The three that matter most, in the order they must merge:
fleet-infra— retire the duplicate Flux render workflowfleet-infra— collapse CI into a single job (must follow the above)workspace— collapse CI into a single job
Those three alone remove the large majority of the recurring spend.
Acceptance criteria
- A workflow run in a private repo allocates a runner and executes at least one step
- The route taken is recorded here — raised limit, or which PRs were bypassed and why
- The three PRs above are merged in the stated order
- Whether a spending limit was raised is stated explicitly, so the next person knows whether the estate is now spending real money
Blocked by
- None — can start immediately, and nothing else in this set can finish until it does.
Activity
ExtraToast commentedon Aug 29, 2026
Capacity restored — a monetary budget was added, and private-repo runs now allocate runners and execute real steps.
All 16 pending pull requests are merged, plus three found along the way (#41, homelab-collections#14, and test fixes on the two fleet-infra PRs). Verified per-run cost in production:
nix-config keeps three jobs because
Nix Flake Checkis a reusable-workflow call that cannot be inlined, and its aggregator has to gate it.Measurement of the actual monthly figure stays open on #39.