research-protocol

name	research-protocol
description	Conduct rigorous research with proper citations (DOI, arXiv, PMID) and source triangulation. MUST BE USED when user asks: "what is SOTA", "recent developments", "compare X vs Y", "is it true that", "research says", "latest papers on", "scientific evidence", "studies show", "state of the art", "literature review", "find papers", "academic research", "benchmark results", "who published", "when was X released", "current best", "what does the research say", "evidence for", "peer reviewed". Searches multiple sources, evaluates reliability, states confidence level. NOT for verifying API signatures (use anti-hallucination) or general web search (use WebSearch directly).
allowed-tools	["WebSearch","WebFetch","mcp__fetch__fetch","mcp__huggingface__get-paper-info","mcp__huggingface__get-daily-papers","mcp__huggingface__search-models","Read"]

Research Protocol

Research Workflow

1. SCOPE    → Define question precisely
2. SEARCH   → Multiple sources, cross-reference
3. EVALUATE → Assess source reliability
4. SYNTHESIZE → Combine findings coherently
5. CITE     → Provide proper attribution
6. ASSESS   → State confidence level

Source Hierarchy

Priority	Source Type	Reliability	Citation Format
1	Peer-reviewed papers	Highest	[Author et al., Year] arXiv:XXXX
2	Official documentation	High	Docs [Library] vX.X
3	Conference proceedings	High	[Conf Year] Paper Title
4	Established tech blogs	Medium-High	[Org] Blog (Date)
5	GitHub repos with citations	Medium	GitHub [repo]
6	Stack Overflow (verified)	Medium	SO [answer-id]
7	General web content	Low	Mention skepticism

Search Strategy

For SOTA (State of the Art)

1. mcp__huggingface__get-daily-papers()  → Recent papers
2. WebSearch("[topic] SOTA 2024")
3. WebSearch("[topic] benchmark comparison")
4. mcp__huggingface__search-models(tags: "[task]")

For Specific Papers

1. WebSearch("paper title arxiv")
2. mcp__huggingface__get-paper-info(arxiv_id: "XXXX.XXXXX")

For Library/API Questions

1. mcp__context7__resolve-library-id() first
2. Then research protocol for context

Citation Formats

Academic Papers

[Author et al., Year] "Title" - arXiv:XXXX.XXXXX
[Author et al., Year] "Title" - DOI:10.XXXX/XXXXX
[Author et al., Year] "Title" - PMID:XXXXXXXX

Documentation

According to [Library] v[X.X] documentation: ...
According to [Framework] docs (2024): ...

Web Sources

According to [Source Name] (Date): ...
Per [Organization] blog (Month Year): ...

Confidence Assessment

Confidence	Criteria	Response Style
HIGH	3+ concordant sources, peer-reviewed	State as established fact
MEDIUM	1-2 reliable sources	Add "according to [source]" caveat
LOW	Conflicting sources	Present multiple views
UNKNOWN	No reliable sources	"I don't know"

Handling Conflicts

When sources disagree:

## Divergent Viewpoints

**Position A** (Source 1, Source 2):
[Description]

**Position B** (Source 3):
[Description]

**Assessment**: [Which seems more credible and why]

Output Format

## Summary

[Main findings - 3-5 lines max]

## Details

[Expanded information with inline citations]

## Sources

1. [Citation 1 with link/DOI]
2. [Citation 2 with link/DOI]
3. ...

## Confidence Level: [HIGH/MEDIUM/LOW]

[1-2 sentence justification]

## Limitations

- [What couldn't be verified]
- [Potential biases in sources]
- [Recency concerns if applicable]

Red Flags (Requires Extra Scrutiny)

• Single source only
• Source older than 2 years (for fast-moving fields)
• Preprint without peer review
• Corporate blog with potential bias
• No citations in the source itself
• Contradicts well-established knowledge

Research Triggers

Automatically engage this protocol when user asks about:

• "What is the current SOTA for..."
• "Recent developments in..."
• "Compare X vs Y"
• "Is it true that..."
• "What does the research say about..."
• "Latest papers on..."
• Scientific claims without citation