Skip to content

Project Roadmap

This roadmap reflects the current post-cleanup state of the repository.

Current State

Recently completed: - repo governance cleanup around main - removal of AI auto-fix and auto-doc workflows from the main merge path - browser extension popup visual polish pass - browser extension shared client/runtime contract refactor - cleanup of stale remote branches

Near-Term Priorities

1. Browser extension hardening

  • add tests around the shared client and runtime contract
  • expand course-context features that use modules, files, and announcements
  • reduce any remaining direct background/runtime coupling

2. SDK agent tools

  • 18-tool agent registry is now live (PR #83)
  • calendar adapter implemented
  • remaining SDK surface area documented in the codebase

3. Docs and site maintenance

  • keep MkDocs pages aligned with current architecture
  • update screenshots/diagrams when UI or structure changes
  • keep workflow/governance pages consistent with real repo settings

4. GemmaReranker isolation

  • isolate GemmaReranker and related data from core app concerns
  • clarify what is core product code versus experimental/support tooling

Delivery Model

small scoped branch
  -> PR to main
  -> CI checks
  -> maintainer review
  -> squash merge
  -> docs site auto-deploys from main

Extension Roadmap Snapshot

The extension is no longer just a future placeholder.

Implemented now: - popup UI - background worker - IndexedDB cache - shared Canvas client layer - shared runtime contract

Still needed: - broader feature coverage - deeper tests - clearer auth strategy if OAuth replaces token entry

Medium-Term Follow-Ups

  • richer extension parity with the TUI
  • improved architecture diagrams reflecting the browser-side layers
  • possible isolation of optional tooling/data into better-scoped directories or repos

Success Criteria

The project is in a good state when: - main stays the only durable branch - docs site reflects actual architecture, not stale plans - extension changes land through a predictable contract layer - maintainers do not need ad hoc branch-protection surgery for normal work