· via dev.to (home feed)
Why tags pushed with GitHub Actions' GITHUB_TOKEN never trigger release workflows
A dev.to post explains how GitHub Actions' built-in GITHUB_TOKEN silently suppresses workflow triggers, so eleven bot-pushed tags produced zero release runs — and outlines fixes that need no new secrets.

What happened
A developer writing on dev.to recounts a release pipeline that quietly did only half its job for two weeks. One workflow ran on every merge to the main branch: it computed the next version, created a v-prefixed tag, pushed it and finished green. A second workflow, release.yml, was supposed to react to that tag, build artifacts and publish a GitHub Release.
Eleven tags went out. Zero release workflows ran. The Actions tab showed neither failures nor skipped runs — nothing at all, as though the tags had never been pushed.
A rule that fails silently
According to the post, the root cause is a deliberate GitHub design decision: events created by the built-in GITHUB_TOKEN never start new workflow runs. Branch pushes, tag pushes, opened pull requests, published releases, labels and issues all become invisible to every other workflow's on: block. GitHub does this to prevent recursion — a workflow that runs on push and also pushes commits would otherwise retrigger itself forever. Only two events are exempt: workflow_dispatch and repository_dispatch, which still fire when sent with GITHUB_TOKEN.
The nasty part is that nothing fails. The push lands, the tag exists and the event is recorded in the repository; it is simply never delivered to the workflow scheduler. The logs show the new tag and the job ends with a green check.
The hidden culprit inside actions/checkout
The author's bump workflow never named GITHUB_TOKEN explicitly. The token leaked in through actions/checkout, which by default authenticates with it and persists those credentials in the local git configuration. Any later git push in the same job therefore runs as github-actions[bot] using that token. The fastest diagnostic, the author suggests, is to push the same tag by hand from a laptop: if the release workflow fires immediately, this rule is the cause.
Collateral damage: pull requests
The same rule bites harder elsewhere, the post notes. Pull requests opened by a workflow using GITHUB_TOKEN never trigger required status checks, so automated dependency-bump PRs can sit forever on "Waiting for status to be reported" because CI was never asked to run. The workflow_run trigger does still fire — it keys off the finished upstream run, which a human push started — but the downstream job sees the upstream commit rather than the tag, so the version has to be recovered from context.
Fixes that require no new secrets
Two remedies avoid adding credentials:
- Reusable workflows: convert release.yml to accept workflow_call with a tag input, then invoke it from the bump job with the tag passed explicitly. Two details matter: a called workflow runs with the caller's github context, so github.ref still points at the branch rather than the tag, and the called workflow cannot hold more GITHUB_TOKEN permissions than the caller grants — the caller needs contents: write if the release uploads assets.
- Dispatch: add workflow_dispatch to release.yml's triggers, grant the bump job actions: write plus contents: write, then run gh workflow run release.yml with --ref pointing at the new tag and GH_TOKEN set to github.token. The release run then checks out the tag and sees refs/tags/ exactly as a manual push would, and it appears as its own run in the Actions tab.
When only a different token will do
If the event itself must trigger things outside your control — required checks on a bot-opened pull request, for example — no amount of workflow chaining helps. The post recommends a GitHub App installation token, minted per run with actions/create-github-app-token and handed to checkout so that every later push in the job uses it. Such tokens expire after an hour and are not tied to any person's account. A personal access token also works, but the author ranks it as the worst option of the set.
Why it matters
This failure mode produces no error, no log line and no red badge — the strongest signal that something is wrong is an absence. Any team whose automation pushes tags, opens PRs or publishes releases with the default token is exposed to silently skipped pipelines and permanently stalled checks. The fix is cheap and structural: chain workflows explicitly through workflow_call or dispatch instead of relying on events the platform deliberately suppresses, and reserve stronger tokens for the narrow cases where an event must genuinely propagate.
- #github-actions
- #ci-cd
- #devops
- #release-automation
- #github