xAI's Grok Bot 72-Hour Build Distilled Into Open-Source Agent Rules
A free MIT-licensed pack distills a 72-hour xAI Grok Bot livestream build into drop-in agent rules, 69 role playbooks, and a failure log.
Takeaways
- Free MIT repo distilling a 72-hour xAI Grok Bot livestream build into reusable agent rules and playbooks.
- Core artifact is AGENTS.md, drop-in house rules a coding agent reads before acting.
- Includes 69 agent role files and nine role-specific playbooks spanning engineering, PM, sales, support, and marketing.
- ANTIPATTERNS.md logs 40 live-stream failures, each paired with the general rule it produced.
- Enforces proof-at-PR: screenshots, before/after numbers, or reproductions required for every change type.
- Hard human-gate list covers auth, payments, migrations, deploys, and any approach that failed twice.
Three engineers from xAI’s Grok Bot team spent 72 hours building and shipping a product during a livestream. The independently maintained field-notes repository converts that session into reusable instructions, role definitions, and failure reports for coding-agent workflows.
The free, MIT-licensed pack matters because it captures operational details that demonstrations often omit: how an agent proves its work, when it must stop, which decisions require human approval, and how a failure becomes a durable rule. Teams can copy individual documents into an existing setup without adopting a new framework.
Inside the repository
The central artifact is AGENTS.md, a file placed at a repository’s root to give compatible coding agents project-level instructions before they act. Support for this convention depends on the agent and tool, so teams should confirm how their environment discovers instruction files.
| Artifact | Contents | Intended use |
|---|---|---|
AGENTS.md |
Rules for planning, implementation, verification, and escalation | Repository-wide agent instructions |
| PDF guide | A 24-page summary of the operating model | Onboarding and reference |
ANTIPATTERNS.md |
Forty livestream failures and the rule derived from each one | Guardrails and review checks |
agents/ |
References on verification, orchestration, skills, routines, and prompts | Designing agent workflows |