source-curation

name

source-curation

description

Build a curated reading list, source bibliography, or recommended-resources digest on a topic via the Tavily MCP. Use when the user wants to learn about a topic from scratch, build a reference collection, share a reading list with their team, or says "build me a reading list on", "what should I read about", "curate sources on", "recommended resources for", "syllabus for", or "starter pack on". Returns a categorized list of high-quality sources (foundational, recent, contrarian) with one-line annotations. For one-shot synthesis, use deep-research instead.

Source Curation

Curate a reading list on a topic — categorized, annotated, and opinionated about what's worth reading first.

When to use this skill

• "Build me a reading list on …"
• "What should I read to get up to speed on …"
• "Curate sources on …"
• "Starter pack for someone new to …"
• "Send the team a syllabus on …"

For a single synthesized report, use deep-research. For tracking fresh news, use news-monitor. For pre-meeting briefs, use lead-research.

Workflow

1. Clarify the level: who is the audience and how deep do they want to go? Beginner / practitioner / expert? 5 sources / 15 / 30+?
2. Cast a wide net: tavily_search with multiple sub-queries:
   • "{topic} introduction" — beginner explainers.
   • "{topic} {year}" — current state.
   • "{topic} survey OR overview OR foundational" — anchor pieces.
   • "{topic} criticism OR critique OR debate" — heterodox voices.
   • For technical topics: "{topic} paper" with include_domains: "arxiv.org,acm.org,ieee.org".
3. Score each candidate:
   • Author / publisher credibility.
   • Recency (especially for fast-moving fields).
   • Depth match to audience.
   • Whether it adds a different perspective vs. items already in the list.
4. Categorize into sections (template below).
5. Annotate: one line per source — what it adds, why it's on the list. Don't re-summarize the article.

Output template

# Reading list: {Topic}
*Curated {YYYY-MM-DD} • Audience: {beginner / practitioner / expert}*

## Start here (foundational, ~30 min total)
1. **{Title}** — {Author}, {Publisher}, {Year}. *{Why it's first}.* [{url}]
2. …

## For practitioners (~2–4 hr total)
3. **{Title}** — {Author}. *{What it teaches}.* [{url}]
4. …

## Recent ({last 12 months})
5. **{Title}** — {Author}, {Publisher}, {Date}. *{What's new here}.* [{url}]
6. …

## Contrarian or critical
7. **{Title}** — {Author}. *{What it pushes back on}.* [{url}]
8. …

## Going deeper (advanced / academic)
9. **{Title}** — {Author}, {Journal}, {Year}. *{What it adds}.* [{url}]
10. …

## Audio / video (if relevant)
- **{Podcast / talk}** — {host / venue}, {duration}. *{angle}*. [{url}]

## Updated periodically
- This list reflects what's available as of {YYYY-MM-DD}. The {topic}
  space moves fast / slow — re-curate every {N} {weeks/months}.

Curation principles

• Quality over quantity. 5 great sources beat 30 mediocre ones.
• Heterogeneous viewpoints. Include something the user will disagree with — that's how you actually learn a topic.
• Mix formats. Long-form essays, papers, talks, short posts. The user's attention budget varies.
• Annotate honestly. If something is "controversial but important," say so. If something is "the canonical reference but dated," say so.
• No dead links. If you can't tavily_extract it successfully, don't include it.

Saving and sharing

Default to saving the reading list as reading-list-{slug}-{YYYY-MM-DD}.md in ~~document store. If the user wants to share with their team, offer to post it to ~~chat (with the link to the doc, not the full content).

Failure modes

• Topic is too broad ("AI"): ask the user to narrow before curating. Otherwise you'll produce a generic list.
• Topic is too niche / new: say so. Curate what exists; flag where the gap is. Don't pad with marginal sources.
• Paywalls: include high-quality paywalled sources but mark them (paywalled) so the user can decide.