KomarDB
testing

Youtube Summarizer

Self-hosted YouTube knowledge-management app (Capture & Understanding Engine)

scores

progress57%
prioritymedium
complexity5/10
readiness3/10
excitement7/10
hours invested2h
last6 days ago

claude · 1 hour ago

Active testing-stage TypeScript app with recent checkpoint progress (4 items ticked) and a brief accidental flip to "live" then back. Momentum is real but launch readiness (3) lags progress (57%), suggesting unclear definition of "done" for v1.

  • Define explicit v1 launch checklist for self-hosted deploy
  • Write Docker compose + README quickstart for one-command install
  • Dogfood on 20 real YouTube videos, log failure modes
  • Decide scope cutoff: Capture vs Understanding engine for v1
  • Fix stage toggle bug that flipped live→development in 3s

notes

Komarlab project

Plan from /cl

  1. Capture & Understanding Engine (Telegram -> transcript -> extraction)
    Forwarding a YouTube link to the bot yields stored light + deep extractions · done 6 days ago
  2. Installable Artifacts -- copy-to-Claude cards + discovery page
    Deep extraction populates artifacts; /library/installables lists them with a Copy button · done 6 days ago
  3. Production deploy -- yt.komarlab.com, Cloudflare Access, webhook live
    App reachable at yt.komarlab.com, UI gated, Telegram webhook registered and healthy · done 6 days ago
  4. Paste-ready deep extraction -- no fabrication, research-first prompts
    copy_prompts never invent commands; mcp vs skill correctly classified · done 6 days ago
  5. Dogfood: validate one-click copy -> paste on real videos
    Copying an artifact works pasted into Claude with no edits
  6. Sub-project B -- semantic search / chat over the library
    Query saved videos by meaning and get cited answers
  7. Sub-project D -- Telegram digest of forgotten items
    Bot proactively resurfaces unreviewed saved items on a schedule

Recent checkpoints

  • 6 days ago1.5hConfirmed the app was already live in production (the apex domain had masked it). Made deep-extraction copy prompts accurate and safe to paste: no fabricated install commands, research-first when the video lacks details, and correct MCP-vs-skill classification. Fixed a latent retry bug that 400'd the Anthropic API whenever a deep extraction failed first-pass validation.