Content Curator Research

You are an autonomous AI research agent. Run every step in order. Do not ask questions. If a source fails, log it and continue.

Config:

{
  "email": "",
  "repo_url": "",
  "repo_clone_dir": "",
  "output_subdir": "content-curator",
  "cadence": "0 6,12,17 * * *",
  "topics": ["LLM", "Claude", "OpenAI", "MCP", "ADK", "agent", "open-source AI", "AI orchestration", "transformer", "fine-tuning", "RAG", "agentic", "multimodal", "reasoning model", "harness", "vibe coding"],
  "star_history_url": "https://star-history.com/compare"
}

> Setup: Replace `, , and ` with your own values before using this skill.

Step 1 — Prepare repo

Expand `` to absolute path. If it doesn't exist:

git clone  

If it exists:

cd  && git pull --rebase origin main

Ensure /content-curator/ exists. Determine today's filename: research-MMDDYY.md (e.g. research-042026.md).

Determine sweep number: if the file doesn't exist → sweep 1. If it exists, count existing ## Sweep headings in it → that count + 1 is this run's sweep number.

Step 2 — HackerNews

WebFetch https://hacker-news.firebaseio.com/v0/topstories.json → first 30 IDs. For each, WebFetch https://hacker-news.firebaseio.com/v0/item/{id}.json. Filter by topic keywords (case-insensitive on title/url). Sort by score. Keep top 5. Extract: title, score, comment count, URL, 2–3 sentence "why it matters" note.

Step 3 — GitHub Trending

WebFetch https://github.com/trending?since=daily and https://github.com/trending/python?since=daily. Filter repos whose name or description matches any topic keyword. Keep top 5 unique. Extract: repo name, description, stars-today, URL, 2–3 sentence summary.

Step 4 — Google AI Blog

WebFetch https://blog.google/technology/ai/rss/. Parse 8 most recent items. Extract: title, pubDate, link, 2–3 sentence editorial summary.

Step 5 — ArXiv

WebFetch https://arxiv.org/search/?searchtype=all&query=llm+agents&start=0 and https://arxiv.org/search/?searchtype=all&query=AI+orchestration+agents&start=0. Top 5 by recency, dedup by title. Extract: title, first 2 authors, full abstract, URL, "why this paper matters" note.

Step 6 — X/Twitter Signals

WebSearches:

• "LLM agents" site:x.com OR site:twitter.com
• "Claude" OR "OpenAI" "agent" site:x.com
• "MCP" OR "ADK" "AI" site:x.com

Extract 5–8 posts: handle, snippet, engagement, 1–2 sentence angle note.

Step 7 — Star History

WebFetch https://star-history.com/compare. Note velocity observations. If none, write "No data extracted."

Step 8 — Synthesize "What's Hot" for this sweep

Generate 4–6 editorial bullets: cross-source themes, surprising signals, narrative threads, angles for builders/PMs/devs. Each bullet becomes an anchor link to a deep-dive section.

Step 9 — Write or append to daily file

Path: /content-curator/research-MMDDYY.md.

If the file does not exist, create it with this top-of-file header before adding the sweep:

# AI Intel Report — MM/DD/YY

> Pick a topic from any sweep's **What's Hot** below to draft content via `/content-curator`.

---

Append this sweep block to the file:

## Sweep N/3 — HH:MM UTC

### What's Hot Right Now (Sweep N)

- [🔥 Theme 1 one-liner](#sN-theme-1-slug)
- [🔥 Theme 2 one-liner](#sN-theme-2-slug)
- [🔥 Theme 3 one-liner](#sN-theme-3-slug)
- [🔥 Theme 4 one-liner](#sN-theme-4-slug)

### Deep Dive (Sweep N)

#### Theme 1 — [full title]
**Why it's hot:** [2–3 sentences]
**Signals:**
- [signal with link + data point]
- [another signal]
**Angle for a post:** [1–2 sentences]

#### Theme 2 — [full title]
(same structure)

(repeat for each theme)

### Raw Signals (Sweep N)

#### HackerNews
| Title | Score | Comments | URL | Why it matters |
|-------|-------|----------|-----|----------------|

#### GitHub Trending
| Repo | Stars Today | Description | URL | What it does |
|------|-------------|-------------|-----|--------------|

#### Google AI Blog
- **[Title]** ([Date]) — summary — [URL]

#### ArXiv Papers
- **[Title]** ([Authors]) — why it matters — [URL]
  Abstract[full abstract]

#### X / Twitter Signals
- @[handle]: "[snippet]" — angle: [note]

#### Star History Notes
[observations]

---

Important: use sN-... prefixes (s1-, s2-, s3-) on anchor names so later sweeps in the same file don't collide.

Step 10 — Commit and push

cd 
git add content-curator/research-MMDDYY.md
git commit -m "research: sweep N/3 for MM/DD/YY"
git push origin main

If push fails, keep the file and log the failure — do not delete.

Step 11 — Print run summary

Sweep complete — MM/DD/YY HH:MM UTC (sweep N/3)
File: content-curator/research-MMDDYY.md (appended)
Pushed: [yes | no — reason]
Items: HN=[N] GitHub=[N] Blog=[N] ArXiv=[N] Twitter=[N] StarHistory=[ok|no data]
Failures: [none | source: reason]

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: msdakot
• Source: msdakot/ai-foundary
• License: MIT

Install and usage instructions live in the source repository linked above.