Craft in your tools.
One memory, every tool: the profile you teach in Claude is the same one ChatGPT drafts from, Cursor edits with, and your CI enforces. Steal a prompt and start where you already write.
- 1Claude
The conversational desk.
The full loop as conversation — drafting, checking, teaching. In Claude Code, the same tools run in your terminal, next to your docs and release notes.
Copy this prompt“Draft this week's issue from these three links in my Newsletter voice, check it, and fix anything that fails.”
Then teach“Approve the final — the closing line is exactly my register.”
- 2ChatGPT
The same memory, over there.
Enable the connector in a chat and ChatGPT drafts from the profiles Claude taught. Switch models freely — your voice doesn't reset.
Copy this prompt“Turn yesterday's issue into a launch tweet — my Tweets profile, check it before you show me.”
- 3Cursor
Writing that lives in a repo.
Docs, marketing pages, changelogs: the agent drafts with compose, gates with check, and keeps product terminology consistent via the glossary — while editing your files.
Copy this prompt“Rewrite docs/quickstart.mdx in our Docs voice and make sure it passes the house checks.”
- 4API & CI
The gate in your pipeline.
The same key drives a plain REST API — so the editorial gate runs where no human is watching. Batch-check up to 50 documents: style violations annotate the PR, policy violations block the merge.
POST /api/craft/check/batch # fail_on=policy → style hits: warnings on the PR → policy hits: the build fails
- 5Pick one
Start where you already write.
Run the loop on one real piece in the tool you used yesterday. The others inherit everything the moment you open them.
The one-prompt start“Set up Craft for me: create a profile called My Voice, import my archive from myblog.com, arm the anti-slop pack, then draft something and show me the check.”