Brief builder
Free, working Claude skills I use to run an AI automation studio. Drop-in SKILL.md files. By Robin Laires / Laires Labs.
npx -y skills add robdasi/skills --skill brief-builderAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Set up a site's voice and its anti-repetition memory once, so every later article sounds human and never recycles a story, a source, or an opening. Captures the business, builds a voice profile (defaults plus a client-voice override and a lexicon), and starts two banks: the E-E-A-T stories used and the stat sources used. Run first, once per site. The other content-pipeline skills load its output. Produces a site brief, a voice profile, and the banks, then stops.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
3.6 KB, as published. Nobody here has run it
Brief Builder
Two things make AI content obvious: it sounds like nobody, and across a site it repeats itself, the same opening, the same client story, the same stat, over and over. This skill fixes both at the source, once per site, so the writing step inherits a voice and a memory of what's already been used.
Run this first. Its output is loaded by every later step.
Inputs (ask for whatever is missing)
- The domain and what the business sells, in one line (required).
- Two or three pages that already sound right, to sample the real voice.
- Optional: the audience in plain terms (who actually reads this), and any existing articles, so the banks start populated.
The method
-
Capture the business once. What it sells, to whom, the reader in plain terms (write for the non-expert buyer, not for SEO professionals), and the real buyer questions they'd type. Keep it as a record the other steps reuse, never re-derived per article.
-
Set the voice defaults. Write like you talk, read it aloud, and if it sounds stiff rewrite it. Vary sentence length on purpose (short 5-8 word lines mixed with 12-18). Never start three sentences in a row the same way. Paragraphs of 2-4 sentences. Opinionated expert sharing what actually works, "I" and "you", like explaining to a smart friend over coffee. Define a technical term on first use. Contractions always. Regular dashes, never em dashes. American English by default, UK spelling for buyer-facing local articles.
-
Build the anti-detection blocklist. Ban the tells outright: filler ("it's important to note", "when it comes to", "in today's digital landscape", "it's worth mentioning"), hedging ("perhaps", "it could be argued"), transition clichés ("let's dive in", "without further ado", "in conclusion"), "Overall," / "Ultimately," paragraph starters, and repetitive triple-structure lists.
-
Layer the client voice over the top. If client samples exist, read two or three and mimic the rhythm, vocabulary, register, and contraction habits. Build a small lexicon (prefer, avoid-and-replace, capitalisation, product names). Where the client voice conflicts with a default, the client wins.
-
Start the E-E-A-T story bank. List the personal or client experiences available to use, and which articles have already used which. The rule downstream: one unique story per article, never reused. Anonymise clients to an industry description, never a name.
-
Start the source bank. List the stat sources already cited and how many articles use each. The rule downstream: every article cites at least one source nobody else has used, and no source appears in more than two articles.
Output
Produce site-brief.md (business, reader, buyer questions), voice-profile.md (defaults, blocklist, the client lexicon and override), and banks.md (E-E-A-T stories used, stat sources used with counts). Write them so later skills load them as context and check against them.
Then stop. Set once per site. Every article this month reads it and stays out of its own ruts.