Skip to content

ROADMAP.md

Latest commit

 

History

History
65 lines (46 loc) · 2.82 KB

ROADMAP.md

File metadata and controls

65 lines (46 loc) · 2.82 KB
Raw

Roadmap

This roadmap is intentionally short. WebCodex 0.2.x is about making online AI coding safer to try and easier to review, not promising a full IDE or autonomous operations platform.

0.2.x: Productize The Online Coding Loop

  • Simpler first-run setup.
  • ChatGPT MCP and GPT Actions acceptance tasks.
  • More fixture dogfood tasks.
  • Clearer review and rollback workflow.

LSP Phase 1: Completed

  • Agent-side supervisor.
  • Constrained rust-analyzer profile.
  • Read-only symbol navigation.
  • Startup capability awareness.

LSP Phase 2 Read-Only MVP: Completed

  • Disk-backed full-text document refresh (didOpen / didChange), without editor-style incremental sync.
  • Bounded document_diagnostics with explicit freshness and timeout semantics.
  • Normalized, bounded hover.
  • Workspace-only symbol search with project-relative results.
  • ToolRuntime, MCP, generic GPT Action, OAuth, schema, and startup capability synchronization.

LSP Phase 3 Multi-Language: Completed

  • Language registry (lsp/language.rs): one LanguageProfile per server owns extensions/languageIds, project markers, executable resolution, default_args, and the constrained read-only initialize profile; the supervisor and navigation handlers carry no per-language branches.
  • Python (pyright) and TypeScript/JavaScript (typescript-language-server) added alongside Rust, validated end-to-end against a real pyright.

Future LSP Work

  • Fan workspace_symbols across every detected language in a polyglot project (currently the first detected language answers).
  • A multi-language start_coding_task.semantic_navigation startup summary (today it is a Rust-focused readiness hint; the runtime tools are already multi-language).
  • Broader diagnostics fidelity beyond the constrained read-only profiles.
  • Explicitly designed write capabilities such as rename and code actions (not part of the read-only MVP).
  • More languages (each is one registry profile plus its read-only safety profile and tests).

Validation Intelligence MVP: Completed

  • Deterministic structured extraction from bounded safe validation metadata.
  • Bounded cargo-check diagnostics and cargo-test failed-test evidence.
  • Conservative validation failure-kind classification without root-cause inference.
  • Shared finish/handoff evidence plus the read-only validation_summary session query.
  • ToolRuntime, MCP, generic GPT Action, OAuth, and schema synchronization (75 runtime tools, 25 GPT Action operations).

Future Coding Intelligence

  • Richer multi-language validation adapters.
  • Optional machine-readable Cargo JSON integration.
  • Review and rollback UX.

Later

  • Dashboard.
  • Ops packs.
  • Browser/computer-use evidence.

Non-Goals

  • Full IDE replacement.
  • Autonomous DevOps by default.
  • Arbitrary computer use as a core promise.
  • Guaranteed compatibility with every AI client.