blog-topic-interview
Pre-writing interview that captures the author's real opinions, experiences, and stances on a blog topic before the post is written on their behalf — the companion that runs BEFORE seo-blog-writer. Us
SKILL.mdskills/blog-topic-interview/SKILL.md
GitHubBlog Topic Interview
A post published under someone's name should say what that person actually thinks. This skill collects the things nobody else can supply — their stance, their real stories, their numbers, their boundaries — before any drafting starts, and files them where every later post can find them.
It is the interview command of who-let-the-blogs-out, packaged on its own. The rules are not paraphrased here; they are the same files.
Run it
- Load
references/governing-rules.md,references/state.md,references/reporting.md, andreferences/evidence-rules.md(its Verification standards set fact expiry dates). - Follow
references/interview.mdfrom top to bottom, with the one override below.
Everything in interview.md applies as written: at most three questions at a time, never re-ask what the bank already answers, every opinion-bank entry carries provenance, numbers go to facts.json, and the packet lands at .blog/posts/<slug>/packet.md.
The one override: no .blog/ yet
interview.md says that when .blog/ is missing you should point at brand. This skill doesn't have brand, so create the same starting tree that brand creates, and nothing more. These files are added to interview.md's declared writes for this run only.
Settle two slugs first, lowercase and hyphenated, and show both in your first message, ahead of the first batch of questions. A slug confirmation doesn't count toward the three-question limit.
- Client slug from the business name:
Cursive Media→cursive-media. Use the domain only when there's no name. It appears in every path and is never renamed. - Post slug from the topic: short, no stop words,
how-often-pressure-wash-driveway. Normallyplanreserves this; here you do.
Create the tree once the slugs are confirmed:
.blog/
├── registry.json {"version": 1, "clients": [<client row>], "posts": []}
└── clients/<slug>/
├── brand.md frontmatter + the seven headings from state.md
├── opinion-bank.md assets/opinion-bank-template.md, copied whole
├── facts.json {"version": 1, "facts": []}
└── notes.md a single `# Notes — <name>` heading
- Client row:
slug,name,domain(ornull),brandProfile: "missing",bankEntries: 0,updatedset to today. brand.md: frontmatter withclient,name,domain,sitemap: null, andupdated. Leave outsourceandtaxonomyuntilbrandruns; they record how the profile was built, and nothing has been yet. Under each of the seven headings, write**Not captured.** Ask in the next brand pass.Never fill a brand section from guesses about the business.- Leave
posts/alone untilinterview.mdcreates the post folder. Fields onlyplanorbriefwould know staynulluntilbrieffills them:primaryKeyword,intent, andcurrentVersioninpost.json, plustitleon its registry row.
If .blog/ exists but this client doesn't, create clients/<slug>/ with the same four files and add the client row. Create the client folder whole, never file by file.
In the closing report, add one line saying the brand profile is empty and that who-let-the-blogs-out's brand command fills it if they adopt the full system.
Never write into this skill's own folder. assets/opinion-bank-template.md is only ever copied out. A bank saved inside a skill is lost when the skill is updated, and it carries one client's private positions into every other project.
Names that point outside this skill
The shared files mention other who-let-the-blogs-out commands. Where one says to hand off to a command this skill doesn't have, stop and name the next step. Don't improvise it.
briefandwrite→ theseo-blog-writerskill runs both, or who-let-the-blogs-out when the user is working in it.brand,plan,verify,images,review,revise,publish,refresh,export→ only in who-let-the-blogs-out.
Hand off
When the packet is written, the next step is the brief and draft. Say that seo-blog-writer picks up from .blog/posts/<slug>/packet.md.
The files under references/ and assets/ are copies of who-let-the-blogs-out's and are checked against it in CI. Change the originals, never these.