research-and-summarize— agent skill

Overview

Information overload is the default state. This skill transforms any research task into a structured summary: headline insight first, context second, detail third, action last. Designed for decision-makers who need clarity, not comprehensiveness.

When to Use

• Summarizing technical documentation or papers
• Researching a technology choice
• Briefing a team on a topic
• Distilling a long document for a specific decision

Process

Step 1: Define the Research Question

1. State the specific question being answered: "Should we use Kafka or RabbitMQ for our event pipeline?"
2. State who the answer is for and what decision it enables.
3. This scopes the research — don't gather information beyond what the decision needs.

Verify: Research question is specific enough to have a clear answer.

Step 2: Gather and Evaluate Sources

1. Identify 3–5 high-quality, authoritative sources.
2. For each source, note: recency, authority, potential bias.
3. Cross-reference key claims across sources.
4. Flag conflicting information — don't silently pick one side.

Verify: Key claims are supported by at least 2 independent sources.

Step 3: Write the Layered Summary

1. Headline (1 sentence): The single most important insight.
2. Key findings (3–5 bullets): Supporting evidence for the headline.
3. Context and nuance (1–2 paragraphs): Caveats, tradeoffs, conditions under which the headline doesn't hold.
4. What we don't know: Gaps in the available information.
5. Recommended action: Given the findings, what should the reader do next?

Deliver: A structured summary with all 5 sections.

Step 4: Cite Sources

1. Every factual claim is linked to a source.
2. Include the date of each source (recency matters in fast-moving fields).

Verify: Every claim has a citation.

Common Rationalizations (and Rebuttals)

Verification

• [ ] Research question defined before research begins
• [ ] Key claims cross-referenced across 2+ sources
• [ ] Summary has: headline, findings, context, unknowns, action
• [ ] Every factual claim has a citation with date

References

• think-before-coding skill
• idea-to-spec skill