I wanted one clean workflow: send a Slack message and end with a merged PR, using a self-contained VM so the whole process stayed safe, stable, and repeatable.

The story in one run

We started by moving the workflow to a dedicated VM. That gave us isolation from my daily machine and a predictable place to run long tasks. Then we installed OpenClaw, connected model providers, wired Slack, and used that setup to create and refine this blog post through a few short feedback loops.

The important shift was not "AI wrote a draft." The shift was operational: Slack request in, repository changes out, PR updated, done.

The win is not the draft. The win is a repeatable Slack-to-PR loop.

How to set this up (short version)

  1. Create a dedicated VM so the agent runtime is isolated and persistent.
  2. Install OpenClaw and initialize your workspace.
  3. Configure OpenClaw markdown context files: AGENTS.md, SOUL.md, TOOLS.md, USER.md, and IDENTITY.md.
  4. Connect AI providers (Google and/or OpenRouter) so you have a primary model plus fallback.
  5. Install GitHub CLI and authenticate with gh auth login.
  6. Install coding tools you want in the loop (Codex, Claude Code).
  7. Connect Slack app/events so requests can trigger work and status can post back.
  8. Run the flow: Slack request -> OpenClaw edits -> review -> push branch -> PR update.

Starter links

What is next

Next step is a reusable Slack template for content tasks so each request includes goal, tone, and acceptance criteria up front. That should cut revision rounds and keep the loop fast.

OpenClaw AI Tools VM Setup Remote Development Slack Claude Code