Skip to content

Feat/agent harness - #31

Merged
xzneozx96 merged 27 commits into
mainfrom
feat/agent-harness
May 29, 2026
Merged

Feat/agent harness#31
xzneozx96 merged 27 commits into
mainfrom
feat/agent-harness

Conversation

@xzneozx96

Copy link
Copy Markdown
Owner

No description provided.

xzneozx96 and others added 27 commits May 29, 2026 17:07
The companion ends grammar/learning answers with the source YouTube URL, which
previously opened externally on youtube.com. Route those links into our own tip
interface instead, mimicking the Discover flow.

Backend: add resolve_curated_video + GET /api/collection/resolve/{video_id}.
It resolves a video_id to its curated home by checking standalone tips and
scanning the curated tip playlists' membership (reusing get_cached_playlist —
no reverse YouTube endpoint exists). Three-way outcome distinguishes a definitive
"not curated" (404) from a transient fetch failure (503) so an API blip never
misclassifies a curated video as external.

Frontend: extractYouTubeVideoId helper; RecommendedVideoLink resolves a chat
YouTube link to /tips/playlist/<pid>?lesson=<vid> or /tips/video/<id> (new tab),
falling back to the external URL while pending / on 404; session-cached, dropped
on 503 for retry. MessageMarkdown's link override is now always-on so the global
companion surface is covered. CourseSidebar scrolls the active lesson into view
when arriving via ?lesson=.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…le/source

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… resume

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ress accessor

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace LLM-based word splitting with jieba for Chinese lessons. The vocab
LLM call no longer decides word boundaries or romanization — it only fills
meaning + usage for a fixed, pre-segmented token list, so it can no longer
omit or inconsistently split words.

- add jieba dep + segmentation_provider (Chinese only, LLM fallback elsewhere)
- enrich_vocabulary: jieba tokens + deterministic pypinyin, LLM meaning/usage;
  coverage guaranteed (missed words kept blank, never dropped)
- wire into _shared_pipeline; extract_vocabulary retained as fallback
- output shape unchanged so frontend buildWordSpans is unaffected

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When TipProgress.title is empty (legacy records pre-dating title persistence),
look up the UserMaterial by courseId and use its name as the continue-item title.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reasoning steps are now individually collapsible (intended) and share the
'Thinking' label with the trace header, so assert the outer trace exists and
is open rather than requiring exactly one match.
…e item

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@vercel

vercel Bot commented May 29, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
shadow-learn Ready Ready Preview, Comment May 29, 2026 3:10pm

@xzneozx96
xzneozx96 merged commit bea04e4 into main May 29, 2026
3 of 5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant