Deep Planning
A repository-aware planning methodology that generates PRDs with phases, success criteria, escape hatches, context budgets, and skill mappings.
Deep Planning is how Knowledge OS handles complex tasks with 3+ phases. It reads the repository structure, audits available skills, and generates a PRD file with independently verifiable phases. Each phase has an outcome, actions, verification method, and context budget (how many files can be in context simultaneously). Escape hatches define when to stop and ask for help. The PRD integrates with Ralph Execute for stateless iteration. Plans produce direction, not implementation: a sub-agent should be able to execute each phase from the prose description alone.
Where it shows up:
Related Terms
Methodology
Ralph Execute
A stateless iteration protocol where all project state lives in a PRD file, enabling multi-session e...
Skills Layer
Skill
A reusable instruction set that encodes a repeatable process. Skills load context, apply methodology...
Foundation
Knowledge OS
A persistent, file-based operating system for AI that turns Claude Code into a context-aware busines...