publish deploy fails on re-run: duplicate github-pages artifact

[daira]

Symptom

Re-running the failed render-and-deploy job in Deploy Rendered Site
(.github/workflows/publish.yaml) fails at the Deploy to GitHub Pages step
(actions/deploy-pages@v5) with:

Error: Multiple artifacts named "github-pages" were unexpectedly found for this workflow run. Artifact count is 2.

Root cause

The Upload artifact step uses actions/upload-pages-artifact@v5, which
uploads an artifact named github-pages. Artifacts are scoped to the workflow
run, not the attempt, and persist across attempts. So re-running the job
creates a second github-pages artifact for the same run, and
actions/deploy-pages — which requires exactly one — aborts.

The action's own hint ("Please re-run the deployment at a later time") is
misleading here: re-running is precisely what creates the duplicate, so a
plain "Re-run failed jobs" never self-heals. The stale artifact must be deleted
by hand first.

Evidence

Run 28734687994 (the
publish for #1323, protocol spec 2026.7.0) had two github-pages artifacts:

artifact id	created
8090060529	2026-07-05T08:40:03Z (original attempt)
8090483436	2026-07-05T09:36:18Z (re-run)

Deleting both and re-running the job succeeded.

Proposed fix

Make the job self-healing by deleting any pre-existing github-pages artifact
for the current run before the upload. upload-pages-artifact@v5 has no
overwrite input, so a dedicated cleanup step is needed — inserted immediately
before the Upload artifact step:

      - name: Delete any stale `github-pages` artifact from a previous attempt
        uses: actions/github-script@<pin to a v7 SHA>
        with:
          script: |
            const { data } = await github.rest.actions.listWorkflowRunArtifacts({
              owner: context.repo.owner,
              repo: context.repo.repo,
              run_id: context.runId,
              name: 'github-pages',
            });
            for (const artifact of data.artifacts) {
              core.info(`Deleting stale artifact ${artifact.id}`);
              await github.rest.actions.deleteArtifact({
                owner: context.repo.owner,
                repo: context.repo.repo,
                artifact_id: artifact.id,
              });
            }

With that in place, "Re-run failed jobs" works without manual artifact cleanup.

🤖 Claude Opus 4.8

[linear]

COR-1514