escoffier-labs/solos-cookbook/last30days

last30days

Attribution: this is an independent, from-scratch implementation of the recency-research concept popularized by mvanhorn/last30days-skill (MIT). No text or code is shared with that project. For the full-featured original, install mvanhorn's skill.

Research what people are saying about a topic right now, not what ranked well six months ago.

Goal

Survey recent discussion across:

• Reddit
• X
• the broader web

Parse intent first

Extract:

1. TOPIC
2. QUERY_TYPE

Use one of these query types:

• RECOMMENDATIONS
• NEWS
• GENERAL

Search plan

Run 6 to 10 searches from different angles.

Reddit

• site:reddit.com {TOPIC}
• reddit {TOPIC} recommendations
• reddit {TOPIC} discussion

X

• site:x.com {TOPIC}
• {TOPIC} x thread
• {TOPIC} trending

Web

For RECOMMENDATIONS:

• best {TOPIC} 2026
• {TOPIC} comparison
• top {TOPIC}

For NEWS:

• {TOPIC} news 2026
• {TOPIC} announcement
• {TOPIC} update

For GENERAL:

• {TOPIC} 2026
• {TOPIC} guide
• {TOPIC} overview

Prefer a past-month freshness filter when the tool supports it.

Deep dive

Fetch the 3 to 5 most promising results in full.

Prioritize:

• active Reddit threads
• detailed long-form articles
• primary-source announcements
• news articles with specifics

Output shape

Recommendations

• rank the most-mentioned options
• explain why they are being recommended
• cite sources
• note disagreements or controversial picks

News

• list key developments from the last 30 days
• include dates when available
• separate facts from community reaction

General

• summarize major themes
• call out where people agree
• call out where opinion splits

Rules

• prefer specific names, dates, and numbers
• weight recent discussion over stale SEO pages
• flag contradictions
• cite sources with links when possible
• keep the synthesis tight unless the user asks for a deep dive

Anti-pattern

Do not answer with a generic web summary built from one or two results. The whole point is cross-source recency.