deep-research

name	deep-research
description	Structured research methodology for technical investigations. Use when evaluating technologies, comparing solutions, investigating incidents, or any task requiring verified, multi-source information gathering.

Skill: deep-research

A structured methodology for conducting thorough technical research. Produces verified, actionable intelligence from multiple sources.

---

When to Use

• Evaluating a technology, library, or service before adopting it
• Comparing competing solutions for a technical decision
• Investigating production incidents or performance issues
• Gathering requirements from scattered documentation
• Any question where the first search result isn't enough

---

Research Workflow

Phase 1: Scope Definition

Before searching anything, define the research boundary:

## Research Scope
- **Question**: {specific question to answer}
- **Context**: {why this matters, what decision it informs}
- **Constraints**: {budget, timeline, tech stack, compliance}
- **Success Criteria**: {what "good enough" looks like}
- **Time Box**: {max research time before synthesizing}

Phase 2: Multi-Source Search

Use at least 3 independent sources. Never rely on a single source for critical decisions.

Source Type	Best For	Tools
Official docs	API details, configuration, limits	Context7, WebFetch, CDP
GitHub issues	Known bugs, workarounds, community sentiment	WebSearch, GitHub API
Stack Overflow	Common problems and vetted solutions	WebSearch
Release notes	Breaking changes, deprecations	GitHub releases, changelogs
Benchmarks	Performance claims verification	GitHub repos, blog posts
Source code	Ground truth when docs are wrong	git clone --depth 1, Grep

Search Strategy:

1. Start with official documentation (highest signal-to-noise)
2. Check GitHub issues for real-world problems
3. Search for recent (< 6 months) comparisons or benchmarks
4. Clone and read source code when docs are ambiguous

Phase 3: Freshness Filter

Information decays. Apply these freshness rules:

Information Type	Max Age	Action if Older
API documentation	3 months	Verify against source code
Security advisories	1 month	Check CVE databases directly
Performance benchmarks	6 months	Note date, flag as "may be outdated"
Architecture patterns	2 years	Still valid if fundamentals unchanged
Library popularity / stars	3 months	Check current numbers

Phase 4: Cross-Validation

Every critical finding needs corroboration:

Finding → Source 1 confirms → Source 2 confirms → HIGH confidence
Finding → Source 1 confirms → Source 2 contradicts → INVESTIGATE
Finding → Single source only → MEDIUM confidence (flag it)
Finding → No source, inferred → LOW confidence (state assumptions)

Red flags that require deeper investigation:

• Official docs and community reports disagree
• Feature claimed in marketing but absent from changelog
• Benchmark numbers that seem too good (check methodology)
• "Works great" posts from accounts with no other activity

Phase 5: Synthesis

Compile findings into the output template below.

---

Output Template

## Research Report: {topic}

### Executive Summary
{2-3 sentences: what was investigated, key finding, recommendation}

### Findings

#### {Finding 1 Title}
- **Claim**: {what the source says}
- **Source**: {URL or reference}
- **Verified by**: {second source or "single source — medium confidence"}
- **Relevance**: {how this affects the decision}

#### {Finding 2 Title}
...

### Comparison Matrix (if comparing options)

| Criteria | Option A | Option B | Option C |
|----------|----------|----------|----------|
| {criterion 1} | | | |
| {criterion 2} | | | |
| {criterion 3} | | | |

### Quality Score

Rate each dimension 1-5:

| Dimension | Score | Notes |
|-----------|-------|-------|
| **Completeness** | /5 | Were all aspects of the question covered? |
| **Freshness** | /5 | How current is the information? |
| **Source Diversity** | /5 | How many independent sources? |
| **Verification** | /5 | Were findings cross-validated? |
| **Actionability** | /5 | Can a decision be made from this? |

### Open Questions
- {What couldn't be determined}
- {What needs further investigation}

### Methodology
- Sources consulted: {count}
- Time spent: {duration}
- Search queries used: {list key queries}

---

Anti-Patterns

Don't	Do Instead
Trust the first result	Cross-validate with 2+ sources
Spend 2 hours searching without synthesizing	Time-box each phase, synthesize incrementally
Ignore information age	Always check publication date, flag stale data
Copy-paste without attribution	Track sources for every finding
Research endlessly	Define "good enough" upfront, stop when criteria met
Skip official docs and go straight to blogs	Official docs first — they're the ground truth
Assume popular = good	Check issues, maintenance activity, and alternatives

---

Quick Reference: Search Commands

# Clone a repo for local analysis (fast, read-only)
git clone --depth 1 https://github.com/org/repo.git /tmp/repo-inspect

# Search for specific patterns across a cloned repo
grep -rn "pattern" /tmp/repo-inspect/src/

# Check project health signals
gh api repos/org/repo --jq '{stars: .stargazers_count, issues: .open_issues_count, updated: .pushed_at}'

# Get recent releases
gh api repos/org/repo/releases --jq '.[0:5] | .[] | {tag: .tag_name, date: .published_at}'