agentic-digests

Weekly AI research digests generated autonomously every Monday and archived here. Each digest is the exact markdown sent to subscribers that week.

Digests

Folder	Topic	Window
ai-ecosystem/	Claude Code, MCP servers, AI developer tooling	Last 7 days
ai-marketing/	AI marketing tools, video, growth automation	Last 14 days
dtc-marketing/	DTC ecommerce, paid social, retention & LTV	Last 30 days

---

How it works

Three GitHub Actions jobs fire every Monday at 6:30am EDT (10:30am UTC) from louloret/last30days-skill, branch custom.

Pipeline (per digest)

1. Fetch subscriber list
         ↓
2. Research: last30days.py
   Reddit · X · HackerNews
         ↓
3. Synthesize: Claude Haiku
   narrative + GitHub repo sections
         ↓
4. Deliver: Resend → subscribers
         ↓
5. Archive: markdown saved here

Research engine

Each digest runs scripts/last30days.py with a hardcoded plan — a set of weighted subqueries targeting specific topics. The engine searches across three sources in parallel:

Source	How it searches	Weight
Reddit	Semantic full-text across topic-specific subreddits	2.5 (highest signal)
HackerNews	Claude-generated Algolia query (1–3 keywords)	1.0 (baseline)
X	Claude-generated OR phrase query, min_faves:3	0.7 (de-weighted for hype)

Claude Haiku generates the X and HN search queries at runtime based on each subquery's intent, so searches adapt to what's actually relevant that week rather than matching a fixed keyword list.

Synthesis

Raw research output (scored, deduplicated posts) is piped through digest/email_digest.py. Claude Haiku reads the compact research and writes a narrative digest covering:

• Top themes and signals from the week
• Trending GitHub repos (new repos gaining stars fast)
• Top all-time repos by topic (ranked by log10(stars) × recency blend)

---

Source repos

• Research engine: mvanhorn/last30days-skill — public upstream
• Digest workflows: louloret/last30days-skill — fork running the scheduled jobs (branch custom)