What Is Railspike?
Railspike is a prompting framework for structured work with coding agents. It addresses requirements drift, misunderstood intent, unwanted dependencies, and long stretches of agent work that diverge from the expected solution.
Its central concept is the plan-prompt: a plan stored as a file that becomes the agent’s working prompt. Because the plan lives in the repository, developers can inspect and revise it before code is written.
Structured Development Process
Railspike divides development into four stages:
- Brainstorm possible approaches and select a direction.
- Outline the chosen solution at a high level.
- Build a detailed development plan from atomic subtasks.
- Complete the implementation by following that plan.
The framework uses multiple instruction files at the points where they are needed, reducing the chance that important guidance disappears as the model’s context changes. It also includes instructions for actions before, during, and after specific stages.
Persistent and Agent-Agnostic Planning
Plans are stored as searchable plaintext in the project, so they remain available across sessions, context compaction, and changes between coding agents. These records also preserve the reasoning behind a feature, allowing a developer or another agent to review earlier decisions.
The same process has been refined with Cursor, Windsurf, Copilot, Claude Code, and Codex. This consistency helps teams move tasks between agents while retaining shared constraints and established intent.
Supported Work
Railspike has been used for new features, changes to existing features, bug fixes, documentation, and checklists for work spanning multiple cycles. Its planning flow is also intended to make large volumes of agent-generated material easier to read and edit outside the console.
A debug mode is included, and the prompting system can be kept in a repository for synchronization between projects.