Skip to content

TODO.md

Latest commit

 

History

History
31 lines (25 loc) · 3.26 KB

TODO.md

File metadata and controls

31 lines (25 loc) · 3.26 KB
Raw

openlearn TODOs

Active

  • Public release candidate dogfood - run manual-tests/public-release.md with five fresh testers across macOS, Windows, and Linux; close every release blocker before selecting a tag.
  • Quick Learn validation - assign the local-file journey in the public gate and separately dogfood folder and public GitHub flows with non-sensitive study material.
  • Select and publish the Community release - after the human gate passes, choose the version, verify the exact candidate hashes, and push the matching immutable vX.Y.Z tag.

Backlog

  • Per-concept review grading - current due-review prompt records one easy/hard/missed result for the whole session; later support mixed outcomes per concept
  • Source import polish - improve Quick Learn depth scaling for very large folders and repositories; consider private repository authentication separately
  • Provider abstraction - extract the urlopen chat-completion call into a small ModelProvider class so adding Anthropic native / Ollama doesn't require touching prompt logic
  • FMHY integration - contact maintainers for license/permission; see docs/DEPENDENCIES.md

Done (recent)

  • Quick Learn MVP - start a separate adaptive topic from a file, bounded folder, or public GitHub repository with no placement or outline approval
  • v0.5.0 Phase 1 - Rich UI rewrite + platformdirs data dir + migration notice
  • v0.5.0 Phase 2 - Review scheduling polish (status-bar count, menu r shortcut, inline easy/hard/missed, rich due table, /review --due)
  • v0.5.0 Phase 3 - PDF/DOCX/URL import + --scan folder auto-import with checksum dedupe
  • v0.5.0 Phase 4 - Coding drill sandbox (/drill, /check, curated bank, pytest integration)
  • v0.5.0 Phase 5 - CLI review prompt, real per-concept Ebisu SRS, import dedupe, User-Agent, .bak snapshots, config-mask audit, happy-path integration test
  • v0.5.0 Phase 6 - YouTube suggestions (zero-dep ytInitialData parser), /videos + openlearn videos, opt-in suggest_videos, version bump to 0.5.0 + --version
  • v0.5.0 test hardening - pexpect workflow smoke tests + JSON tutor regression evals for observed REPL/tutor behavior failures
  • Menu import parity - new-course + Context-files menus now offer file/URL/folder import via a shared import core; fixed mislabeled .txt options, version-tracked User-Agents, video-suggestion dedup on fetch failure
  • Dry-run prompt preview - chat, resume, next, and review support --dry-run to print rendered prompts without calling the API or mutating local files
  • Session resilience - transient provider failures retry with bounded backoff, in-flight REPL answers survive process restart for Enter resubmission or typed replacement, and openlearn repair fixes simple corrupt JSON frontmatter with .bak backup
  • PyPI release automation - vX.Y.Z tags build and verify distributions, publish through trusted publishing, and create GitHub releases with artifacts
  • Phase 1-4 (v0.4.0): tutor correctness, system-prompt quality, session compression, test coverage
  • v0.4.0: Answer evaluation loop, dynamic prompt, /done UX, review scheduling scaffold, Rich UI groundwork, version bump + tag