When should you use it?
- Support waits on engineers for simple lookups
- Customer questions need account data
- You want an audit trail of agent actions
Not for: Teams unwilling to log agent actions; Write access to production on day one
What do you need, and what do you get?
You need
- Read-only database role
- Billing system read key
- Slack workspace
You get
- Replies with results where the question was asked
- An audit log of every query
How does it work, step by step?
- Create read-only accessGive the agent a database role and billing key that cannot write.
- Keep secrets outsideRoute calls through a gateway that adds credentials so the agent never holds them.
- Start with tools disabledTurn on destructive tools only per tool and only when needed.
- Answer in threadHave the agent post results and files where the question was asked.
- Log everythingKeep an append-only record of each action.
The prompt
Our own prompt, inspired by shivsarthak on Hacker News and written to work in Custom agent and Claude. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
Support question in thread: {question} Using read-only tools only, find the relevant account in {systems}. Post the answer with the query you ran and the rows or records used. If the request needs a write or you are unsure which account, ask in the thread instead.
Inspired by shivsarthak on Hacker News: Hacker News launch comment for Doozy, a Slack agent for support teams. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Prompt injection can still misuse allowed tools; keep them narrow.
- Customer data in Slack threads needs the same care as anywhere else.
Our take
This is a launch post, so read product claims with that in mind, but the security design is worth borrowing even if you build your own bot. No credentials in the sandbox, gateways that add them per request, and read-only queries are the right defaults. shivsarthak's starting point is a real one: support kept waiting on engineers for simple account lookups. Whether or not you use his product, copy the security choices. Give the agent a database role and billing key that cannot write, keep credentials in a gateway that adds them per request so the agent never holds them, and switch on any destructive tool individually only when you truly need it. Answering in the same Slack thread keeps context where the question was asked, and an append-only log of each action gives you an audit trail. Prompt injection can still misuse the tools you allow, so keep them narrow. Customer data shared in Slack deserves the same care as anywhere else, and write access to production should not be on the table at the start.
Where the time goes back: Support gets answers to data questions in minutes instead of waiting in an engineer's queue.
FAQ
Why no credentials in the sandbox?
Who used it most?
Source and credit
Inspired by shivsarthak on Hacker News. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- Hacker News launch comment for Doozy, a Slack agent for support teams
- Creator
- shivsarthak on Hacker News
- Published on
- Hacker News
- Original date
- Oct 20, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Custom agent
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
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.