Last30Days Skill — Deep Dive
A source-backed tour of how Last30Days turns a long SKILL.md contract and a Python engine into a multi-source research system.
10 Chapters
Section titled “10 Chapters” 🧭 1. Product Thesis & Surface Area What the skill is actually building: a social-relevance search engine wrapped as an agent skill. → 📜 2. SKILL.md as Control Plane Why a giant instruction file is not an accident here, and how it coordinates the host model. → 🐍 3. CLI Entry Point & Runtime The thin Python CLI, process cleanup, parser surface, and execution boundaries. → 🧠 4. Query Planning & Topic Resolution How Last30Days decides where to search before it searches. → 🔌 5. Source Adapters & Auth Model Reddit, X, YouTube, web search, GitHub, and the BYO-credentials design. → ⚖️ 6. Ranking, Clustering & Fusion Weighted RRF, reranking, per-author caps, entity-aware clustering, and cross-source story merging. → 🧾 7. Rendering & Shareable HTML How the engine emits a strict model-facing envelope and a separate polished HTML artifact. → 🗃️ 8. Watchlists, Storage & Briefings SQLite persistence, delta tracking, scheduled topics, and daily/weekly briefings. → 🧪 9. Testing & Release Discipline What 90+ test files and release docs say about how the project is run. → 🏗️ 10. Why the Design Works The bigger pattern: one skill, one engine, many sources, and very explicit guardrails. →
Built by LangCraft · Source analysis based on mvanhorn/last30days-skill v3.3.2