Agent Skill
2/7/2026

research

Structured research methodology with scoping, multi-source collection, triangulation, source credibility scoring, and anti-hallucination guards. Use when investigating technologies, comparing tools, evaluating libraries, exploring unfamiliar codebases, or producing research reports.

B
bigpapicb
1GitHub Stars
1Views
npx skills add BigPapiCB/Universal-Claude-Skills

SKILL.md

Nameresearch
DescriptionStructured research methodology with scoping, multi-source collection, triangulation, source credibility scoring, and anti-hallucination guards. Use when investigating technologies, comparing tools, evaluating libraries, exploring unfamiliar codebases, or producing research reports.

name: research description: Structured research methodology with scoping, multi-source collection, triangulation, source credibility scoring, and anti-hallucination guards. Use when investigating technologies, comparing tools, evaluating libraries, exploring unfamiliar codebases, or producing research reports. metadata: version: "1.0" allowed-tools: WebSearch WebFetch Read Grep Glob

Research

Decision Tree

Research request → What depth?
    ├─ Lookup (simple fact, quick answer) → 3-5 sources, skip to Collect + Deliver
    ├─ Synthesis (compare options, summarize topic) → 8-12 sources, all 5 phases
    ├─ Analysis (architecture decision, tech evaluation) → 12-15 sources, all 5 phases + recommendation
    └─ Investigation (novel/complex, no obvious answer) → 15+ sources, all 5 phases + multiple rounds

Phases

Phase 1: Scope → Phase 2: Collect → Phase 3: Validate → Phase 4: Synthesize → Phase 5: Deliver

Phase 1: Scope

Before searching, define the boundaries:

  • Core question — What specifically are we trying to answer?
  • Depth level — Lookup / Synthesis / Analysis / Investigation
  • Search strategy — Keywords, year constraints, domain restrictions
  • Success criteria — What would a good answer look like?

Phase 2: Collect

Execute multi-angle searches. Don't just search once — approach the topic from multiple directions:

Search AngleExample Query Modifier
Factual"what is [topic]"
Comparative"[topic] vs [alternative]"
Technical"[topic] architecture implementation"
Practical"[topic] production experience lessons learned"
Recent"[topic] 2026" or "[topic] latest"
Critical"[topic] problems limitations drawbacks"

Source diversity required:

  • Official documentation (most authoritative)
  • Official blogs and announcements
  • Independent technical analysis
  • Community discussions (GitHub issues, forums)
  • Academic papers (for foundational claims)

Parallel searches: Execute 3-5 searches simultaneously when tools support it.

Phase 3: Validate

Every claim needs triangulation:

Claim from Source A → Find Source B that independently confirms or contradicts
    ├─ Confirmed by 2+ sources → High confidence
    ├─ Contradicted → Note conflict, investigate why
    └─ Only one source → Flag as unverified, note in output

Source chain: Official docs → Independent analysis → Community verification

URL validation: Check that every cited link is accessible before including it.

Phase 4: Synthesize

Score and weight sources using:

FactorWeightHigh ScoreLow Score
Authority30%Official docs, core maintainerRandom blog, no credentials
Recency25%Current year, last 12 months3+ years old
Specificity25%Directly answers the questionTangentially related
Independence20%No vendor affiliationVendor marketing material

Conflict resolution:

  1. Prefer official documentation
  2. Check publication dates (newer often supersedes)
  3. Note the disagreement explicitly
  4. Present both perspectives if unresolved

Phase 5: Deliver

Output structure (scale to depth):

DepthOutput Format
Lookup1-2 paragraphs + sources
SynthesisSections with comparison table + sources
AnalysisExecutive summary + detailed sections + recommendation + sources
InvestigationFull report with findings, methodology, limitations, sources

Every deliverable must include:

  • Inline citations [Source Name](url) for all claims
  • Sources section at the end with all referenced URLs
  • Confidence indicators for uncertain claims
  • Date of research (findings may become outdated)

Anti-Patterns

Anti-PatternFix
Single-source claimsTriangulate with 2+ independent sources
Hallucinated citationsNever fabricate URLs — only cite pages you actually fetched
Vendor benchmarks as neutralExplicitly note source bias
Overly broad queriesFocus each query on a single specific aspect
Missing recencyAdd current year to search queries
Shallow explorationRead 10+ sources, not just the first 1-2 results
No conflict disclosureAlways note when sources disagree
Broken links in reportValidate every URL before including
Assuming first result is bestCheck multiple results, compare quality
Skipping the scope phaseAlways define what "done" looks like before searching

For detailed source evaluation criteria see:

Skills Info
Original Name:researchAuthor:bigpapicb