When should you use it?
- You publish on Substack and also keep a website or portfolio blog
- Routine edits and cross-posting eat your publishing time
- You can describe exactly where content goes
Not for: Edits you have not written yourself; Accounts where an accidental email blast would be serious and you cannot watch the first runs
What do you need, and what do you get?
You need
- The exact text, links and formatting to add
- A mapping table from site links to newsletter links
- Which posts should and should not be emailed
You get
- Live post updated with paragraphs placed exactly
- A new post migrated with formatting, converted links and subscribe buttons
- A linking note added to an older post
- A written log of each action, failure and fix
How does it work, step by step?
- Write the task specFor each job, state the post, the exact insertion point, the content and whether subscribers get an email.
- Provide a link mapList how every site URL maps to its newsletter equivalent so links convert before content goes in.
- Let it insert and checkThe agent finds the spot, inserts content and takes screenshots to verify layout.
- Confirm the email settingHave it check the publish dialog for the send option before every update or new post.
- Ask for the play-by-playRequest a step-by-step account of what it did, including failed attempts, and review the live posts.
The prompt
Our own prompt, inspired by Prisca Onyx and written to work in Claude. Values in lime are placeholders.
Task for my Substack [publication]: In the published post "[title]", insert the two paragraphs below immediately after the paragraph that contains "[anchor phrase]", keeping all links and italics. Because this edits a post that is already live, do not send it to subscribers; confirm in the publish dialog that no email will go out. After saving, open the live post and verify every link URL and formatting. Then write me a step-by-step log of what you did, including any attempts that failed. [paste content]
Inspired by Prisca Onyx: I Walked Away While Claude Edited My Blog. Then I Asked It to Show Me Everything.. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Its first insertion method merged new text into an existing paragraph; it undid that and switched to the editor's own insert command at an exact position.
- Ask for verification of every link URL and format on the live page, not only in the editor.
Our take
Prisca Onyx handed three Substack jobs to Claude's Chrome extension and walked away: insert paragraphs with links into a live post without re-sending it, migrate a portfolio article with converted links and subscribe buttons, and add a note to an older post. She then had it document every step and failure. Asking it to explain itself is the real lesson. A browser agent that silently succeeds leaves you unable to tell luck from reliability, while a written log of failures and recoveries shows where it can break. The safety detail worth copying is the explicit email decision per task, since a stray newsletter send cannot be recalled. Start with low-stakes edits, keep the link map in a file, and review the live result after each run until you trust the pattern. Confirm audience and publish settings before each migration. Prefer drafts over live edits until the play-by-play looks clean. Back up posts before bulk changes.
Where the time goes back: Precise edits and migrations with many links are fiddly click work; the agent does them while you are away and reports exactly what changed.
FAQ
How does it read the source article?
Did it email subscribers?
Source and credit
Inspired by Prisca Onyx. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- I Walked Away While Claude Edited My Blog. Then I Asked It to Show Me Everything.
- Creator
- Prisca Onyx
- Published on
- priscaonyebuchi.substack.com (Substack)
- Original date
- Mar 5, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Claude
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; nothing was sent, filed, purchased or published.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.