Blog Post Creation Skill

You are helping the user create a new blog post for their Hugo-based blog at dabase.com.

Your Task

1. Determine the current year:

- Run: date +%Y to get the current year dynamically

1. Ask the user for:

- What is the blog post about? (Get a detailed description of the topic/content) - What title do they want? (Or offer to generate one based on the topic)

1. Generate an SEO-friendly title and description:

- Create a concise, compelling title if they don't provide one - Write a brief, SEO-optimized description (1-2 sentences) based on what the blog is about - Confirm both with the user before proceeding

1. Generate the filename from the title:

- Convert to lowercase - Replace spaces with hyphens - Remove special characters (keep only alphanumeric and hyphens) - Example: "My Great Post!" → "my-great-post.md"

1. Create the blog post using Hugo:

- Run: hugo new content/blog/YEAR/[filename].md (where YEAR is from step 1) - Hugo will automatically create the file with the correct frontmatter

1. Update the frontmatter:

- Read the newly created file - Update the title field with the confirmed title - Update the description field with the SEO-friendly description - Keep the auto-generated date field

1. Confirm completion:

- Show the file path: content/blog/YEAR/filename.md - Let them know they can start writing content - Suggest running hugo server to preview if needed

Publishing gotchas — check these when a post won't render

Hugo silently drops content instead of erroring, so "it built fine but the page isn't there" is nearly always one of these two:

1. A date in the future. Compare against date -u, not local time — BST/SGT

local time is ahead of UTC, so a post stamped T12:00:00Z at 11:20 UTC is invisible until noon. Never hardcode a time of day; use the real publish instant. This bit the podcast generator (content/podcast/default.md.do), which now reads YouTube's .timestamp from the metadata cache.

1. A draft: key. We don't use Hugo drafts at all — they slow things down.

Omit the key entirely; don't write draft: false. To keep something out of the feed, don't generate a file for it (see NON_EPISODES in content/podcast/all.do).

1. A podcast episode missing from the YouTube playlist. redo all discovers

episodes solely by enumerating the "AI Infrastructure" playlist (PLiKgVPlhUNuyTXzN03gCB1lqvaHXxPLak) in metadata/episodes.json.do, and episode numbers come from position in it — reversed, so the newest video becomes the highest number. A published-but-unlisted video generates no markdown at all and the build looks like a no-op. Add it to the playlist rather than working around the script.

Changing the podcast generator

The episode markdown and metadata/episodes.json are committed and are redo targets, because GitHub Actions can't run yt-dlp (YouTube rate-limits it) so the site build needs them in the repo. redo won't rebuild a target it has no record of building, .redo/ is gitignored, and redo -f respects that too:

warn 040-benchmarking-ai-infrastructure.md: already existing: not redoing

So editing default.md.do appears to work while silently changing only brand-new episodes. Run redo regen (see regen.do) to delete the derived files and rebuild the whole back catalogue, then review git diff. With no pipeline change a regen is a no-op in git, which is the signal it worked.

After publishing, verify rather than assume — hugo --quiet then confirm the page and, for podcasts, that grep -c '<item>' public/podcast/index.xml matches the number of episode files. CI (.github/workflows/test.yml) enforces that.

Important Notes

• Use date +%Y to dynamically determine the current year
• Use Hugo's hugo new command to bootstrap the post
• Generate SEO-friendly titles and descriptions
• Confirm the title and description with the user before creating the file