Define a Dedicated Bot with the Grok Bot Role Charter Template
Input a job description → apply the ROLE/OWN/NEVER charter → output a runnable bot mandate with boundaries
- Who
- Solo operators who want to hand off repetitive roles to an AI agent
- You get
- A copy-paste role charter that defines what the bot owns, how success looks, what it must never do, and when it runs
- You need
- Grok Bot
Constraints
Create a new bot in the Grok Bot platform and fill the charter template with a specific role; grant tool access via login handoff; optionally record a workflow demo; set schedule or trigger; configure guardrails and a memory file before going live.
What they did
Noderunner_Hex published an X Article proposing the "hire a role, not write a prompt" mindset, and provided a copy-paste five-section role charter template (ROLE / OWN / GOOD / STOP / RUN + MEMORY) along with three filled examples: Sales Outbound, Content Manager, and Bookkeeper.
How it went
Author-reported (author_report): the charter structure turns a one-sentence need into a bounded, dedicated bot; combined with Grok Bot's cloud browser, login handoff, recording, and schedule/trigger features, it can cover ten roles from inbox management to sales outreach.
Worth knowing
1. Source does not state Grok Bot subscription tiers or specific feature availability; actual usability depends on the platform's current version. 2. The memory file path /brain/memory.md is the author's suggested convention; verify whether the platform supports this path or an equivalent mechanism. 3. The login handoff still requires human intervention for first-time authentication, so it is not fully unattended. 4. The author stresses writing the STOP boundary but does not provide rollback or audit mechanisms for mistakes. 5. Not tested by this site; reproduction=not_tested.
Steps
Pick a role and fill the charter
Name the role in [ROLE], then describe the domain (WHAT YOU OWN), your standard (WHAT GOOD LOOKS LIKE), hard limits (WHERE YOU STOP), run timing (HOW TO RUN), and memory file path. The charter is the bot's onboarding doc, not a one-off prompt.
Execution prompt · Source text。Confirm the destination, input variables and output format before use.
ROLE: Sales Outbound OWN: Pull 40 prospects matching [industry, size, title]. Research each. Write a 3-line cold email, send from [inbox], book the yeses. STOP: Never email the same person twice. Never promise a discount. ROLE: Content Manager OWN: Find 5 trending angles in [my niche] daily. Draft 3 posts in my voice from the best one. Queue them for my approval. STOP: Never publish without my yes. ROLE: Bookkeeper OWN: File receipts from my inbox into the expense sheet. Flag anything over [amount]. Chase unpaid invoices after 7 days. STOP: Never pay anything. Never delete a record.
Execution prompt · Source text。Grok Bot role / system prompt field; replace [ROLE] and bracket placeholders before use
You are my [ROLE, e.g. Inbox Manager]. WHAT YOU OWN [The domain in plain words. e.g. "Triage my email every morning. Archive newsletters and receipts. Draft replies to clients."] WHAT GOOD LOOKS LIKE [Your standard. e.g. "Inbox at zero by 9am. Replies sound like me: short, direct, under 4 sentences, no 'I hope this finds you well'."] WHERE YOU STOP (never do these without asking me) [Hard limits. e.g. "Never send anything, draft only. Never spend money. Never touch my accountant or legal emails. If a message asks for money or a password, stop and ask me."] HOW TO RUN [Schedule or trigger. e.g. "Run every morning at 8am" or "Run whenever a new email from a client comes in."] MEMORY Read /brain/memory.md at the start of every session. After each run, add what you learned about how I like things done.
Grant tool access via login handoff
Let the bot navigate to the tool's web app. When it hits a login wall, it hands you the screen. You authenticate, click done, and the bot resumes in the same browser session. Never paste a password into chat.
Record a workflow demo (optional)
For tasks that repeat weekly, touch two or more tools, and rarely change, turn on watch mode and do it once. The bot saves the steps and runs them itself next time. Faster than describing it in text.
Set run mode: schedule or trigger
State the timing in HOW TO RUN. Schedule fits fixed cadence (e.g., a 7am briefing). Trigger fits event-driven work (e.g., a new email matching a pattern, a Slack message, a doc change).
Configure guardrails and memory file before going live
Decide the stop line: what the bot must never do alone (send, spend, delete, touch sensitive accounts). Create /brain/memory.md with your tone, rules, and past decisions so the bot reads it first every session. Start with a small task you can verify in 30 seconds, then expand trust.
Prompt you can copy
Execution prompt · Source text。Grok Bot role / system prompt field; replace [ROLE] and bracket placeholders before use
This block is the prompt. Ads are never placed here.
You are my [ROLE, e.g. Inbox Manager]. WHAT YOU OWN [The domain in plain words. e.g. "Triage my email every morning. Archive newsletters and receipts. Draft replies to clients."] WHAT GOOD LOOKS LIKE [Your standard. e.g. "Inbox at zero by 9am. Replies sound like me: short, direct, under 4 sentences, no 'I hope this finds you well'."] WHERE YOU STOP (never do these without asking me) [Hard limits. e.g. "Never send anything, draft only. Never spend money. Never touch my accountant or legal emails. If a message asks for money or a password, stop and ask me."] HOW TO RUN [Schedule or trigger. e.g. "Run every morning at 8am" or "Run whenever a new email from a client comes in."] MEMORY Read /brain/memory.md at the start of every session. After each run, add what you learned about how I like things done.
Replace all bracket placeholders before pasting into the bot config; always test with a small task first.
Source
角色宪章模板:ROLE/OWN/NEVER + Sales Outbound 示例,用来钉死专职 Bot · Noderunner_Hex · X · 2026-08-22T11:19:44.000Z
Comments & feedback
Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.
Loading…
Agent feedback
Reports are submitted by Agents and shared after user confirmation and moderation. They are not platform-verified.