research-methodology

name	research-methodology
description	How the Research Analyst agent conducts research, evaluates sources, and produces structured reports
license	Apache-2.0

Research Methodology

The Research Analyst agent follows a systematic methodology to produce reliable, well-sourced research reports.

Research phases

1. Scoping

Before any search begins, the agent defines:

• Topic boundaries -- What is in scope and what is not.
• Depth requirement -- Surface-level overview vs. deep technical analysis.
• Output expectations -- Report length, audience, and required sections.

If the user's request is ambiguous, the agent asks clarifying questions before proceeding.

2. Broad search

The agent uses WebSearch with multiple query variations to find a diverse set of sources:

• Rephrase the topic using synonyms and related terms.
• Search for both supporting and opposing viewpoints.
• Look for primary sources (research papers, official docs) and secondary sources (analysis, commentary).
• Target different content types: articles, documentation, forum discussions, code repositories.

3. Deep reading

For the most promising results, the agent uses WebFetch to read full pages and extract:

• Key statistics and data points.
• Direct quotes from domain experts.
• Methodology details (for research papers).
• Publication dates and author credentials.

4. Cross-referencing

Every factual claim is verified against at least two independent sources:

• If multiple sources agree, the claim is marked High confidence.
• If only one source supports it, the claim is marked Medium confidence with a note.
• If sources contradict each other, both positions are presented with context.

5. Synthesis

The agent identifies patterns across sources:

• Common themes and areas of consensus.
• Active debates and unresolved questions.
• Gaps in available information.
• Emerging trends that multiple sources hint at.

6. Report writing

The final report follows a fixed structure: Executive Summary, Key Findings, Detailed Analysis, Sources, and Confidence Assessment. Every factual claim includes an inline citation.

Source evaluation criteria

The agent ranks sources using these factors:

Factor	Weight	Description
Authority	High	Is the author or organization a recognized expert?
Recency	High	When was the content published or last updated?
Evidence	Medium	Does the source cite its own references?
Objectivity	Medium	Does the source acknowledge limitations and counterarguments?
Corroboration	High	Do other independent sources agree?

When to use local tools

If the research topic relates to code in the current project, the agent supplements web research with local analysis:

• Grep to find relevant patterns, imports, and usage.
• Read to examine specific files in detail.
• Glob to understand project structure.
• Bash to run commands like git log for project history.

This is especially useful for questions about the project's architecture, dependencies, or implementation details.