PRD Execution

Multi-session projects that survive context loss

Variable (10-120 iterations)Manual: Impossible without the pattern

The stateless execution pattern for complex projects. A PRD file holds all state: current phase, iteration count, next action, blockers. Each invocation reads state, executes one action, writes state. Zero memory held in context. 42 PRDs, 300+ iterations, zero lost state.

Workflow Steps

1

Plan

/deep-planning

Generate PRD with phases, success criteria, escape hatches

2

Execute

/ralph-execute

Read state, execute one action, write state, signal

3

Verify

/ralph-execute

Check success criteria and quality gates

4

Iterate

/ralph-execute

Continue until TASK_COMPLETE or BLOCKED

Example

A 25-phase SEO domination campaign spanning 3 weeks. Each iteration reads the PRD, performs one action (write an outline, build a page template, deploy), updates state, and signals ready. Session breaks, context compaction, and agent handoffs: none lose state.

Deep Dives

Knowledge OS GuideClaude Code for GTM

Ready to automate this workflow?

Every workflow ships with Knowledge OS. Set up your system in 90 minutes.

View Pricing All Workflows

Built and maintained by Victor Sowers at STEEPWORKS