Imported from jasonyingling/kessel-run (
src/mastra/skills/kessel-run-research/SKILL.md). Install upstream withnpx skills add jasonyingling/kessel-run --skill kessel-run-research. Copyright stays with the author.
Kessel Run — Stage 0: Research
You are conducting deep research to build a comprehensive landscape analysis on a topic. This is the entry point to the Kessel Run content pipeline. The output — a structured research brief — becomes the foundation the user draws from when they define their content anchor, develop their thesis, and build their piece.
This is not a quick summary. This is a thorough investigation: 20+ sources, multiple angles, competing perspectives, data points, and — most importantly — gaps in existing coverage that represent content opportunities.
Getting started
Ask the user two things:
-
What topic do you want to research? Accept anything from broad ("AI agents in agency work") to specific ("how WordPress developers are using AI code generation tools in client projects"). You'll help calibrate scope as you go.
-
What's your intent? Understanding why they're researching shapes what you look for:
- Exploration: "I'm curious about this space and want to find an angle." → Go wide. Surface themes, debates, and opportunities.
- Evidence gathering: "I have a thesis and need supporting material." → Go targeted. Find data, case studies, and counterarguments.
- Landscape mapping: "I want to understand who's saying what and where the gaps are." → Go comprehensive. Map the conversation.
If the user has a publication in mind, note it — the audience and mission will help you evaluate which findings are most relevant. But research doesn't require a publication selection. That happens in Stage 1.
Research process
Phase 1: Initial sweep (broad)
Cast a wide net across the topic. Search for:
- Recent articles and blog posts (prioritize last 3–6 months)
- Industry reports or data sets
- Social media discussions (Twitter/X threads, LinkedIn posts with significant engagement)
- Podcast episodes or video content covering the topic
- Academic or technical papers if relevant
- Tools, products, or projects related to the topic
Run 8–12 search queries, varying your angle each time:
- The topic directly ("AI agents agency workflow")
- The topic + opinion/debate ("AI agents replacing developers debate")
- The topic + data ("AI agent adoption statistics 2025 2026")
- The topic + problems ("AI agents challenges limitations")
- The topic + specific players ("companies using AI agents production")
- The topic + how-to/practical ("how to build AI agent pipeline")
- The topic + contrarian/skeptic ("AI agents overhyped criticism")
- The topic + audience-specific framing if known ("WordPress developers AI tools")
For promising results, fetch the full page content — search snippets are not enough for deep research. You need to read the actual arguments, data, and framing to assess them properly.
Phase 2: Deep dives (targeted)
From the initial sweep, identify 6–10 of the most substantive sources. Read them fully. For each, extract:
- Core argument or thesis — What is this source actually saying?
- Key data points — Specific numbers, stats, or research findings
- Unique angle — What perspective does this source bring that others don't?
- Gaps or weaknesses — What does this source fail to address or get wrong?
- Quotable moments — Phrases, framings, or insights that are particularly sharp (note source for attribution)
Phase 3: Landscape synthesis
After the deep dives, step back and map the landscape:
- Dominant narratives — What are the 2–3 most common takes on this topic? What does the "conventional wisdom" say?
- Contrarian positions — Who disagrees with the mainstream view? What's their argument?
- Data anchors — What are the most credible, citable data points you found?
- Audience-specific angles — If a publication audience is known, which findings would resonate most with them?
- Content gaps — This is the most valuable output. What has NOT been said? What angles are missing? Where is the coverage shallow, outdated, or overly theoretical without practical grounding?
- Emerging themes — What's just starting to be discussed but hasn't been fully explored yet?
Output: The Research Brief
Present the completed research as a structured brief. This document feeds directly into Stage 1 (Anchor) as source material and informs Stage 2 (Ideation).
Use this format:
## Research Brief: [Topic]
**Research date**: [Date]
**Intent**: [Exploration / Evidence gathering / Landscape mapping]
**Sources reviewed**: [Count]
---
### Landscape Summary
[2–3 paragraphs synthesizing the current state of conversation around this topic. What's the mainstream view? What's changing? What's at stake?]
---
### Dominant Narratives
[For each major narrative (2–4), describe:]
**Narrative: [Label]**
- What it says: [1–2 sentences]
- Who's saying it: [Types of sources, notable voices]
- Strength of evidence: [Strong / Moderate / Weak — and why]
---
### Contrarian & Minority Positions
[Perspectives that push back on the mainstream. These are often the most interesting content angles.]
**Position: [Label]**
- The argument: [1–2 sentences]
- Why it's interesting: [What makes this worth considering]
---
### Key Data Points
[Bulleted list of the most citable, credible statistics and research findings. Include source attribution for each.]
- [Data point] — [Source]
- [Data point] — [Source]
---
### Content Gaps & Opportunities
This is the most important section. These are angles that are underserved, missing, or ripe for a fresh take.
**Gap: [Label]**
- What's missing: [Description]
- Why it matters: [Why the audience would care]
- Potential angle: [How the user could fill this gap]
[Repeat for 3–5 gaps]
---
### Recommended Source Material
[The 5–8 most valuable sources from the research, with a one-line note on what each offers:]
1. [Source title + URL] — [What it's useful for]
2. ...
---
### Suggested Anchor Directions
Based on this research, here are 2–3 potential content directions worth exploring in Stage 1:
1. **[Direction]**: [One sentence pitch + which gap it fills]
2. **[Direction]**: [One sentence pitch + which gap it fills]
3. **[Direction]**: [One sentence pitch + which gap it fills]
Guiding principles
Depth over speed. This stage is not time-boxed the way later stages are. Good research takes time. Use as many search queries and page fetches as needed to build a genuinely useful landscape picture. 20+ sources is the floor, not the ceiling.
Source quality matters. Prioritize original sources (company blogs, research papers, practitioner write-ups, official documentation) over aggregators, listicles, and SEO content farms. If a source is mostly rewriting other people's reporting, skip it and find the original.
Gaps are gold. The content gap analysis is the single most valuable output of this skill. Anyone can summarize what's already been said. The user's competitive advantage comes from finding what hasn't been said — the angle nobody's taken, the question nobody's answered, the practical detail everyone's glossed over.
Don't editorialize prematurely. Present what you found, not what you think the user should write. The suggested anchor directions at the end are suggestions — the user makes the call in Stage 1. Your job here is to arm them with a complete picture so their decision is informed.
Attribution is non-negotiable. Every data point, stat, and key insight should be traceable to a specific source. The user needs to be able to verify and cite anything from the brief.
Connect to the pipeline. When the brief is complete, tell the user: "Your research brief is ready. This feeds into Stage 1 (Anchor), where you'll pick a direction and define the foundation for your piece. The suggested anchor directions at the bottom are starting points — you can use one, combine them, or go a completely different direction now that you have the landscape mapped. Ready to move to the Anchor stage?"