Draft writer
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 draft-writerAssembled 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
Write the article to the outline, in the site's real voice, with every number sourced and nothing recycled. Holds the voice defaults and the anti-detection blocklist, front-loads the citable answer, weaves links into real sentences, uses the one unique E-E-A-T story, and never invents a statistic. Run after outline-drafter. This is the method, not a magic writer. Produces a draft and 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
3.3 KB, as published. Nobody here has run it
Draft Writer
The writing is the easy ten percent, but only because the nine steps before it were done. By here you have the gap analysis, the outline, the voice profile, and the banks. This skill writes to all of it, and holds the rules that keep the page human, sourced, and unlike everything else on the site.
Run after outline-drafter, with the voice profile and banks loaded.
The hard rules (do not break)
- No fake data. Every number, percentage, and dollar figure comes from a named, verifiable source from the research. If you don't have the real number, leave it out. No invented chart data, no made-up results.
- One unique E-E-A-T story. Use the experience the outline marked, from the bank, that no other article has used. At least two first-person experience statements in the body, woven in, not forced. Name a client's industry, never the client.
- One source nobody else used. Cite at least one source that none of the site's other articles cite, and don't lean on any source more than two articles do.
- No templated link sentences. Weave links into sentences that would exist anyway. Never "For a full breakdown, see our...", "For a detailed comparison...", or anything starting "For a/the" that points at a blog link.
The method
-
Write in the voice. Apply the voice profile as hard constraints: write like you talk, vary sentence length, contractions, no em dashes, and none of the blocklisted filler or transition clichés. Read a section back, if it sounds stiff, rewrite it.
-
Front-load the answer for AI. Under the first question-form H2, open with the 40-60 word self-contained answer the outline placed. Get at least two cited statistics with named sources into the first third. Name tools, brands, places, and numbers, density of real entities is what gets a page cited.
-
Facts from research, opinions from you. Every statistic, quote, and URL comes from the research phase, never from model memory. The take, the framing, and the experience are yours. That split is what keeps it both accurate and human.
-
Vary everything across the article. Different opening from the site's other pieces, no sentence starter three times running, sections that don't all follow the same shape. Scannable, so a skimmer gets the point from headers alone.
-
Conclude with a CTA, not a summary. No "in conclusion", no recap, no link dump. One action-oriented close.
Output
Produce the draft in markdown, starting with an ## heading (never #), in the site voice, every claim sourced, the E-E-A-T story in place, links woven in. Append a short note: which bank entries it used (story and unique source) so they get marked off, and anything left for a human to verify.
Then stop. Hand to publish-prep for the metadata, schema, and the full pre-publish check.