ricardodantas Hugo Write Post
Write a new blog post for a Hugo site that matches how the author already writes. This skill
learns the author's voice and conventions from their existing posts, delegates the writing to the
social-content companion skill, then places the result as a proper Hugo content file.
Hugo specifics (front-matter formats, hugo new, archetypes, sections, taxonomies, build/serve)
live in references/hugo.md — read it when placing the file or resolving
conventions.
Workflow
0. Preflight — companion skill
The prose is written by the social-content skill. Confirm it's available (it appears in your
available skills). If it's missing, tell the user and offer to install it —
npx skills find social-content — and don't start writing until it's available.
1. Confirm this is a Hugo repo
Look for a Hugo config (hugo.toml, hugo.yaml, hugo.json, or config/…) alongside a
content/ directory (and usually archetypes/). If they're absent, stop and tell the user this
skill only runs inside a Hugo site.
2. Get the topic
Ask the user for the post topic if they haven't given one. Capture any specifics (angle, key points, target length) they volunteer — but do not invent constraints they didn't ask for.
3. Learn the author's voice
Run the analyzer to extract the mechanical conventions across existing posts. Your working directory is the Hugo repo, not this skill — invoke the script by its path inside this skill's directory, pointing it at the repo's content dir:
"<skill-dir>/scripts/analyze_style.py" content # defaults to recent 5
"<skill-dir>/scripts/analyze_style.py" content --recent 8 # sample more posts
It reports the front-matter format and key frequency, the section/dir posts live in, the filename pattern, length/heading/list/code stats, the tag/category vocabulary, and the most-recent files.
If it finds too few posts to learn from (a new or nearly-empty blog), say so and ask the user to point at 2–3 exemplar posts to imitate; if there are none, proceed with clean, conventional Hugo defaults and tell the user you did so.
Then read the 2–3 most-recent full posts it lists to capture the qualitative voice that a script can't: tone, sentence rhythm, opening and closing patterns, formatting habits, how headings and lists are used, first vs third person. Summarize all of this into a short style profile (mechanical conventions + voice) — this is what makes the output read like the author.
4. Write the post with social-content
Hand social-content the topic and the full style profile plus 1–2 representative excerpts,
instructing it to write a complete blog post in that exact voice, structure, and length. It is the
primary writer — don't hand-write the body yourself when it's available.
Write only truthful content: don't invent facts, statistics, quotes, dates, or links to fit the
voice. Where the author would cite something specific, leave a clearly marked TODO: placeholder
for them to fill rather than fabricating it.
5. Place the file with correct front matter
Prefer Hugo's own generator so archetypes and front matter are honored:
hugo new content <section>/<slug>.md # e.g. hugo new content posts/my-topic.md
Match the author's conventions from the style profile: same section/dir, filename pattern, and
front-matter format (TOML +++, YAML ---, or JSON). If hugo new isn't available, mirror the
newest post's front-matter shape by hand. Fill the fields the author actually uses (title, date =
now, tags/categories drawn from their existing vocabulary, description, slug). Default draft: false (publishable) unless the user asks for a draft. Then write the body from step 4.
See references/hugo.md for the details.
6. Style-fidelity check
The post must read exactly like the author. Compare the draft against the style profile:
- Front-matter keys/format match the author's posts
- Word count within the author's typical range
- Heading density, list/code usage, and paragraph length in the author's habit
- Tone, person, opening/closing pattern match
- No invented facts, quotes, or stats — any specifics are real or left as marked
TODO:placeholders
If anything is off, send it back through social-content with the specific gap and revise. Repeat
until it matches.
7. Preview (optional)
Offer to render it so the user can check it:
hugo server -D # live preview incl. drafts
hugo # one-off build
Report the created file path and the local URL.