Skip to content

Latest commit

 

History

History
33 lines (23 loc) · 1.65 KB

File metadata and controls

33 lines (23 loc) · 1.65 KB

CodeSharp Roadmap

Goal

Make CodeSharp feel at least as capable and trustworthy as Claude Code in day-to-day coding work.

Phase 1: Reliability and Product Honesty

  • Only surface REPL commands that actually work end to end.
  • Add regression tests for session compaction, tool-call ordering, provider request translation, and interrupt handling.
  • Fix server/session correctness issues, especially session identity and event delivery.
  • Replace fragile prompt-only behaviors with runtime invariants where possible.

Phase 2: Semantic Code Intelligence

  • Implement real LSP lifecycle management instead of the current stub manager.
  • Add symbol-aware tools for definition lookup, reference search, diagnostics, and project-wide symbol navigation.
  • Feed semantic context into the agent automatically so fewer turns are wasted on regex search and file paging.
  • Use diagnostics and symbol graphs to validate edits before returning control to the user.

Phase 3: Agent Workflow Quality

  • Add task-specific validation loops: build, test, lint, and targeted reruns after edits.
  • Improve edit safety with transaction previews, undo support, and clearer multi-file change summaries.
  • Add stronger long-context handling so large repo analysis does not degrade into context-limit failures.
  • Support durable subtask execution and richer plugin-backed workflows.

Immediate Execution Order

  1. Product honesty in the REPL help/completion surface.
  2. Session and provider regression tests for compaction and tool-call protocol correctness.
  3. Fix HttpServer/SessionStore session ID and event-stream behavior.
  4. Replace the LSP stub with a working symbol/diagnostic pipeline.