When should you use it?
- You check the agent every few minutes and burn tokens re-asking
- A routine job is stable enough to widen autonomy on a calendar
- You want interrupt rules written into the profile on day one
Not for: Jumping to week-four autonomy on a brand-new high-stakes job; Removing evidence requirements as you widen autonomy; Ignoring exceptions the bot escalates
What do you need, and what do you get?
You need
- The routine job and its done artifact
- What counts as an exception
- Which actions always need approval even in week four
You get
- A four-week handover schedule
- A standing interrupt rule for the profile file
- A weekly summary format for week four
How does it work, step by step?
- Week 1Drafts only; nothing leaves; you read everything.
- Week 2You clear each action yet skip rereading every step.
- Week 3Routine cases proceed; only odd cases escalate.
- Week 4Scheduled runs continue; you skim the week rollup.
The prompt
Our own prompt, inspired by Carnage (@0xCarnagee) and written to work in Grok Bot, ChatGPT agent, Claude and others. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
Write a four-week trust ramp for this routine job: {job}. Done artifact: {artifact}. Exceptions that always escalate: {exceptions}. Actions that always need my approval: {always_approve}. Produce: Week 1 — drafts only, nothing leaves, I review all of it. Week 2 — I clear each action; skip narrating every step. Week 3 — routine cases proceed; escalate only exceptions. Week 4 — keep the cadence; I skim a week rollup with links. Day-one standing rule: ping me only when you need an approval, lack data, or hit out-of-scope work. Otherwise complete the work and present it. Keep evidence requirements in every week.
Inspired by Carnage (@0xCarnagee): X Article: The Grok Bot Team's Own Workflow (11 steps). The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Breaking runs to re-ask costs tokens and prevents context from compounding.
- Write the interrupt contract into the profile on day one.
- Efficiency claims in quotes are speaker anecdotes, not your baseline.
- Keep evidence tapes even when you stop watching every step.
Our take
Carnage treats trust as a month-long habit, not a vibe: week one full review, then scheduled sample checks, then wider scope only after clean runs. Micromanaging forever and trusting overnight are both failure modes. The calendar is what makes the ramp honest. In week one you read every draft and nothing leaves; in week two you approve actions without rereading each step; by week three only odd cases come back to you; by week four you skim a weekly rollup. Moving forward should depend on clean runs, not on impatience, and a brand-new job with real stakes should start at week one again. Writing the interrupt rule into the profile on day one tells the bot exactly when to stop and ask, which saves the tokens you would burn re-asking mid-run. Keep the evidence even after you stop watching each step, and keep human approval on anything that goes out. Treat efficiency numbers from other people as anecdotes, not your baseline.
Where the time goes back: You replace constant check-ins with a calendar of widening autonomy, so attention moves to exceptions instead of routine steps.
FAQ
Can I skip to week three?
What is the standing rule?
What do I read in week four?
Source and credit
Inspired by Carnage (@0xCarnagee). We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- X Article: The Grok Bot Team's Own Workflow (11 steps)
- Creator
- Carnage (@0xCarnagee)
- Published on
- X
- Original date
- Aug 19, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Grok Bot
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. Calendar ramp; interrupt rule; approvals retained.
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 24, 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.