ROADMAP.md

Roadmap

This roadmap keeps the project focused on practical harness engineering instead of feature sprawl.

Near-term

• Improve the Ralph Loop preset with more realistic runner templates for Codex-style non-interactive runs
• Add optional project-type presets such as:
  • batch-processing
  • repo-coding
  • research-collection
  • UI-validation
• Add richer validator templates without making them default-heavy
• Improve generated tasks.json and feature/task manifest patterns

Medium-term

• Add cross-platform scaffold examples for PowerShell, Bash, and Python runner styles
• Add optional evaluator and grader presets for high-risk tasks
• Add browser-verification guidance bundles for UI-focused harnesses
• Add sample generated projects under an examples/ release artifact or separate branch

Long-term

• Distill newer harness doctrine as model capabilities evolve
• Remove scaffolding that becomes unnecessary as base models improve
• Support narrower presets instead of growing one giant generic scaffold
• Keep the project publishable as a compact open-source skill, not a heavyweight framework

Non-goals

• Becoming a universal agent runtime
• Replacing OpenHarness, Ralph, or upstream harness frameworks
• Defaulting every project to multi-agent orchestration
• Encoding every project-specific habit into the public skill