A label on a GitHub issue now fires the event setup that took a couple of days by hand

The parse step couples two artifacts. An Actions workflow lifts event title, date, region, campaign name and target audience out of the issue body as text ^[9][11]. The form template and the parser have to agree on the shape of that text, and issue forms do not version that agreement. Rename a field in the form and the workflow searches for a string that is no longer there. Issue forms constrain what a human types, which is most of the benefit ^[9].

The trigger is a label, not the submission. Each workflow opens with a condition that runs only when the label is present ^[10]. A human decides when the machinery starts, which is what you want for an event that can still be cancelled after the issue is filed. A label can also be removed and re-added. Every workflow sitting behind one has to be safe to run twice.

Of the six stages in her runbook, five run once per event and one repeats every morning until the event date: download the registrant list, clean it up, post a status update ^[4][18]. Automation pays hardest there, because the execution count scales with lead time and not with the number of events. She wrote that the tasks are easy individually but "an opportunity to paste the wrong link, skip a day, or misspell a campaign name that 15 downstream reports depend on" ^[5].

The build order matters more than the toolchain. "So I didn't write the code. I wrote down my runbooks, handed them to GitHub Copilot, and grew the automation in conversation," she wrote ^[6]. The written runbook was the input. Copilot can only generate a step someone can describe in sequence. GitHub's marketing teams already kept one issue per project, so the plan and the discussion were in the repository before any of this ran ^[8].

"API or CLI, the requirement is the same: a scriptable way in," she wrote ^[15]. The CRM is the interesting case. Authentication went through a browser sign-in instead of a key she had to configure ^[14], and a browser-based login does not usually mix with an unattended runner. Whether that CLI step runs inside Actions or from her own machine is not in the post. The difference matters most for the post-event CRM upload ^[4].

On whether a packaged marketing platform would have covered this, she is specific: APAC is a collection of very different markets, and the same webinar may run in Japanese for Tokyo one month and in Korean for Seoul the next, with different segments, different CRM fields and a different definition of a good lead ^[16]. It is also the test for whether the pattern transfers: where the variation is small and the fields are stable, a vendor's field model absorbs it and there is no parser to maintain.

The account is one practitioner's, published on GitHub's own blog, without the workflow code, run counts or failure rates ^[19]. She limits her own claim to repetitive work that runs through tools offering an API or a CLI ^[17].