Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

Skunkworks Sloth: a set of agents automating my bookmarks review, keeping me focused and growing my portfolio and skills

Description and use

A trio of AI agents automates curation of 8,000+ bookmarks, generates daily project cards, and publishes prototypes to GitHub, reducing manual effort to a few Discord replies.

Problem

Managing over 8,000 bookmarks became impossible to review manually, hindering learning and project initiation.

Outcome

The agent swarm reduced manual bookmark review to three daily Discord replies, automatically curated links, generated daily project cards, and produced prototype repositories on GitHub, enabling focused learning and portfolio growth.

Workflow

  • Curator reads all bookmarks, removes dead links, and sorts remaining links into build, read, and Valhalla buckets.
  • Coordinator turns curated build links into two‑hour project cards, asks weekly for focus preferences, offers cards daily, and hands one reading link per day.
  • Worker receives a selected card, builds the prototype within a two‑hour box, documents learning, and waits for user approval before pushing to GitHub.
  • User interacts via Discord: taps a card, sends "go", and later "approve"; reads daily link and reviews code.
  • Coordinator updates a GitHub Projects Kanban board and a second brain in OKF format.
  • Worker can choose model (OpenRouter or Anthropic) for building tasks.
  • All agents run in separate Hermes profiles and communicate through dedicated Discord channels.

Evidence

I took a sample of bookmarks from my Google Chrome (which currently holds ~8,130 bookmarks).

Source

The Curator reads the whole list, drops the dead links and sorts the rest into build, read and Valhalla (the removal bucket).

Source

The Coordinator turns the build pile into two‑hour project cards, asks me once a week what I care about, offers a few cards every morning, and hands me one link a day from the reading pile.

Source

The Worker builds the card I pick inside a hard two‑hour box, writes down what it learned, and waits for my approval before anything reaches GitHub.

Source

  • Record date: 2026-10-26T05:11:28.620Z