Research

What this skill does

# Research

A general-purpose internet research skill for producing well-sourced, structured responses across any domain. Uses WebSearch and WebFetch to find and evaluate authoritative sources. This skill is not for codebase exploration or project-level investigation — use the Explore agent or Grep/Glob tools for that.

Prioritize authoritative and primary sources over aggregated, ad-heavy, or clickbait content.

## Core Workflow

### 1. Analyze the Question

Before searching, classify the research request:

- **What is being asked?** Extract the core question. Restate it precisely.
- **What type of answer fits?** Select a response template (see Template Selection Guide below).
- **What domain does this fall in?** (science, engineering, policy, history, medicine, law, etc.)
- **What depth is needed?** Quick factual lookup vs. deep synthesis.

### 2. Plan the Search Strategy

Reason about where the best information lives *before* searching broadly:

- **Primary sources first.** For a given topic, identify where the original/authoritative information is published. Examples: official documentation for software, journal articles for science, RFC specs for protocols, government databases for policy/statistics, court opinions for law.
- **Search specific sites.** Use `site:` operators or domain-restricted searches to target authoritative sources identified above.
- **Broaden only if needed.** If targeted searches don't yield sufficient results, widen the search incrementally.
- **Evaluate as you go.** Apply the quick source heuristics below to each result before incorporating it.

### 3. Evaluate Sources

Apply these quick heuristics to every source. For deeper evaluation, consult `references/source-evaluation.md`.

**Trust signals (favor these):**
- Institutional domains (.edu, .gov, .org for established nonprofits)
- Peer-reviewed journals and conference proceedings
- Official project documentation and specifications
- Named, credentialed authors with institutional affiliations
- Content with citations, references, and methodology descriptions

**Distrust signals (avoid or deprioritize):**
- Heavy ad load, clickbait headlines, listicle format ("Top N...")
- Content farms, SEO-optimized filler, affiliate marketing sites
- Anonymous or uncredentialed authors
- No citations, no methodology, no references
- Outdated content presented as current

**Rule of thumb:** The more ads on a page, the less trustworthy the content.

### 4. Select and Follow a Response Template

Choose the template that best fits the research question. Each template contains a checklist and structured outline to follow.

## Template Selection Guide

### Factual / Definitional
**Template:** `assets/templates/factual.md`

**Use when:** The question asks "What is X?", "Define X", "Explain X", or seeks a clear, sourced explanation of a concept, term, entity, or phenomenon.

**Do not use when:** The question is comparing alternatives, asking for steps/procedures, or surveying a broad topic across multiple sources.

**Examples:** "What is WebAssembly?", "Define herd immunity", "What does the Commerce Clause cover?"

---

### Comparative Analysis
**Template:** `assets/templates/comparative.md`

**Use when:** The question compares two or more options, technologies, approaches, or positions. "X vs Y", "How does X compare to Y?", "Which is better, X or Y?"

**Do not use when:** The question is about a single topic (use Factual) or asks for a process (use Procedural).

**Examples:** "PostgreSQL vs MySQL for time-series data", "Compare FPTP and ranked-choice voting", "REST vs GraphQL"

---

### How-To / Procedural
**Template:** `assets/templates/procedural.md`

**Use when:** The question asks how to accomplish something, requesting steps, instructions, or a procedure. "How do I...", "What are the steps to...", "How to set up..."

**Do not use when:** The question is conceptual (use Factual), comparative (use Comparative), or diagnostic (use Troubleshooting).

**Examples:** "How to set up SSH key authentication", "How do I file a FOIA request?", "Steps for PCR amplification"

---

### Literature Review / Survey
**Template:** `assets/templates/literature-review.md`

**Use when:** The question requires synthesizing multiple sources on a topic, identifying consensus, disagreements, and gaps. "What does the research say about...", "Survey the literature on...", "What are the main perspectives on..."

**Do not use when:** A single authoritative source can answer the question (use Factual), or the question is about a specific procedure (use Procedural).

**Examples:** "What does the research say about remote work productivity?", "Survey perspectives on universal basic income", "What's known about long COVID mechanisms?"

---

### Troubleshooting / Diagnostic
**Template:** `assets/templates/troubleshooting.md`

**Use when:** The question involves diagnosing a problem, identifying root causes, or resolving an issue. "Why is X happening?", "How to fix...", "X isn't working..."

**Do not use when:** The question is about understanding a concept (use Factual) or following a known procedure (use Procedural).

**Examples:** "Why is my Docker container exiting immediately?", "Diagnose intermittent network timeouts", "Why did my sourdough not rise?"

---

### Current State / Landscape
**Template:** `assets/templates/landscape.md`

**Use when:** The question asks about the current state, recent developments, or trajectory of a field, technology, or issue. "What's the state of...", "What's happening with...", "Where is X headed?"

**Do not use when:** The question is about a stable, well-defined concept (use Factual) or comparing specific alternatives (use Comparative).

**Examples:** "What's the current state of quantum computing?", "Where is the EV market headed?", "What's happening with CRISPR regulation?"

---

### General (Fallback)
**Template:** `assets/templates/general.md`

**Use when:** The question requires sourced research but doesn't fit any of the templates above. The response needs a flexible structure adapted to the specific content.

**Do not use when:** One of the six specific templates above is a reasonable fit. Prefer a specific template over General whenever possible — structured templates produce more consistent, thorough results.

**Examples:** "What are the implications of X for Y?", "Explain the relationship between X and Y", "What should I know about X before doing Y?"

---

### Multiple Templates

Some research questions span multiple template types. When this happens:

- Identify the primary question type and use that template as the backbone.
- Incorporate relevant sections from secondary templates where they add value.
- Note which templates were combined in the response structure.

## Additional Resources

- **`references/source-evaluation.md`** - Detailed source quality rubric with scoring criteria, domain-specific guidance, and edge cases. Consult when source credibility is ambiguous or contested.