The Agent-Powered Repo Playbook

What this means for you

Petrova is a playbook that turns a scaffolded repo into one a Claude Code session can pick up cold and drive end-to-end. If you ship with AI agents and you’re tired of re-explaining the project every Monday, this is the missing layer.

The pitch

A small pack of prompts and templates. Paste the bootstrap prompt on day one and the agent generates the three files that hold the rails: a workspace map, a phase schedule, and a delegation grammar. After that, every fresh session reads three files and knows exactly what to do.

Who it’s for

The solo founder or small team building real products with Claude Code who keep losing context between sessions and want their agents to wake up oriented.

Proof points

• Three load-bearing files, not a framework: CLAUDE.md, MILESTONES.md, AGENTS.xml
• Five prompts cover the whole loop: bootstrap, phase open, phase close, verification round, drift check
• Currently keeping 31 repos in one workspace coordinated under the same playbook
• Generates no application code — petrova writes the control loop, you write the product

mindmap
  root((Petrova))
    Bootstrap
      one prompt, day one
      reads your spec docs
      asks 7 calibration questions
    Three files
      CLAUDE.md
      MILESTONES.md
      AGENTS.xml
    Five prompts
      phase open
      phase close
      verification round
      drift check
    Not a framework
      no build target
      no runtime
      methodology only