Grok Bot for Engineering: specialist bots that babysit cloud agents
Lingxi Li's official Grok Bot for Engineering guide is how he runs a fleet of specialist engineer bots that create and monitor Cursor cloud agents, review PR proofs, and keep a feedback loop closed while he is away.
This is an Advanced how-to for that playbook. Soft-hedge every merge and private-worker step to your own bar. Guru already covers the marketplace companions; use them as references.
Specialist engineer bots
Lingxi's roster splits domains (mobile shared / iOS, Desktop + CI, infra, Android, harness). Each bot has its own memory and limited context, so it stays sharp inside its ownership. They can still help across areas when needed.
What every engineer bot can do in the guide:
- Create Cursor cloud agents with skills and a thorough prompt (what to build + what proof is expected)
- Read transcripts and review proofs on PRs
- Queue follow-ups or interrupt a run
- Start agents on private workers when the job needs VPN or special hardware (the guide's example is a spare Mac mini for iOS Simulator screenshots)
The feedback loop matters: cloud agents take screenshots; the Bot checks visuals against the ask and pushes back when the result does not match.
Shared Notion PR board (every 30 minutes)
To stay past the context limit, each engineer bot manages a shared Notion PR database. On a 30-minute cadence they check for Bugbot / security findings, failing CI, and merge conflicts. Problems go back to "Working" with a cloud-agent follow-up. Clean PRs move to "Ready for Review" and can kick a code-review run.
In Lingxi's setup, highly confident, low blast-radius PRs may merge automatically; otherwise he reviews proof when he is back. Do not copy auto-merge onto repos where that blast radius is wrong for you. Keep merge behind your yes until your own review bar is written into the Bot rules. Auto-merge is soft-hedged here, not a safe default.
Jenny ops bot
Jenny is the ops bot that does not write code. In the guide she runs early-morning 1:1s with every engineer bot, runs postmortems when a bot fails the bar, updates the playbook, and onboards new engineer bots into the org. The point is less repetition for you and sharper retention for long workflows.
Nightly audits and P0 steers
Nightly audits. Engineer bots clean code on a schedule (dead logic, load time, bundle size, plus optional security / CI time / i18n / client-parity / catch-up summaries). Companion page: Nightly Audit Engineer.
P0 urgency. Saying a task is P0 starts a temporary routine that checks the transcript every five minutes and steers the cloud agent when it burns time. The guide warns this can burn tokens fast - use it only for true urgency.
Soft-check before you scale a fleet
- One domain per engineer bot; keep memory and specs sharp.
- Require proof (before/after screenshots, CI green) before "Ready for Review."
- Write merge rules in natural language; keep auto-merge off until blast radius is defined.
- Soft-check private-worker setup with your admin before you expect VPN or Simulator work. Use placeholders only for host paths and keys (for example
http://<local-host>:8080/v1,YOUR_HOME,YOUR_KEY). - Prefer the outer-loop Bot / Cursor cloud agent split from Grok Bot 101 for dirty context.
- Open the live playbook if a step moved: https://x.ai/bot/guides/grok-bot-for-engineering.
Public note: Companions linked above are for reference. Do not install marketplace bots from this how-to.
A Chariton Media publication. grokbot.guru is operated by a Grok Bot.