Nothing to hand? — an engineer's real-looking research notes about an offscreen-culling trick in a canvas library, with the code, the benchmark, and the approach that failed first. Or try the thin-notes example to see the honest “no story here yet” verdict.
Paste the notes
Code excerpts, “why” comments, decisions, numbers. The free readiness scan counts the raw ingredients — words, code, measurements, “why” language — before you run anything.
The AI finds the story
First the angle: what would go wrong without this, the key insight, the surprise. If the material has no story, it says so and lists the research questions to answer — instead of padding out filler.
Draft, checked
A full markdown draft that frames the problem before the solution, graded against a seven-point checklist. The app recounts the words and cross-checks the outline against the actual headings.
Derived from the @tldraw/write-tbp skill and the tldraw blog and writing guides on SkillSafe.
What does Blog Forge produce?
One JSON result rendered as five things: the angle (the problem, the key insight and the surprise), an outline, the full blog post as markdown, a seven-point self-evaluation checklist with a note on every item, and a list of next steps. The post downloads as a markdown file with YAML frontmatter — title, date, description, topic, audience and word count — so it drops straight into a Hugo, Jekyll, Astro or Eleventy content folder without retyping anything.
What happens if my notes are too thin for a post?
It tells you, instead of padding. The verdict is weak, no draft is written, and the next steps become the specific research questions you need to answer before there is a post worth writing. The “Try thin notes” button shows this outcome without spending anything.
How do I know the draft did not invent things?
The app checks the model rather than trusting it. Every file path and every measurement in the draft is matched back against your notes, and anything that appears in the post but not in your notes is listed for you to verify. The word count is recounted in the browser and compared against the target for the length you asked for, the outline is checked against the draft's real headings, and the draft is re-scanned for the AI writing tells the prompt bans — “Moreover”, “seamless”, “delve”, “leverage”, trailing “, allowing you to” phrases and hedging. Disagreements are shown under “The app double-checked the model”.
Can I fix one part of a draft without redrafting everything?
Yes. Every failed checklist item gets a “Fix this” button, and there is a free-text revise box. Either one sends the existing draft back with your instruction attached, so the model revises what you asked about and leaves the rest alone. A revision is one run, charged like any other, and afterwards the app shows which checklist items were fixed and which regressed compared with the previous draft.
What costs credits and what is free?
The readiness scan, the markdown rendering, the grounding and tone checks, the run-to-run comparison and every export run in your browser and cost nothing. Only drafting and revising call the model. The reserved amount is shown before you run, and the run button is disabled with the exact shortfall if your balance will not cover it. The two bundled examples replay saved runs, so you can see both a strong result and a weak one without spending anything.
Are my past drafts kept?
Past drafts are stored against your SkillSafe account and mirrored into this browser, so they follow you to another device rather than living in one browser's storage. Reopening a past draft also puts the notes that produced it back in the form, which is what makes a second pass possible.