dozybot001

web-research

Conduct web research, gather information from multiple sources, and synthesize findings. Use when task involves searching, comparing sources, or aggregating information. Covers process, synthesis techniques, and output structure. Note: MAARS Task Agent uses provided artifacts and general knowledge—no live web search.

dozybot001 7 Updated 3mo ago
GitHub

Install

npx skillscat add dozybot001/maars/web-research

Install via the SkillsCat registry.

SKILL.md

Web Research

Guidelines for research tasks that involve gathering and synthesizing information from multiple sources.

MAARS Context

Live web search available: Use WebSearch to find current information, benchmarks, and documentation. Use WebFetch to retrieve full page content from URLs for citations. Combine with ReadArtifact for dependency outputs. When citing, include the URL from WebSearch/WebFetch results in your References section.

Research Process

  1. Define scope: Clarify what information is needed, from which domains, and what format the output should take.
  2. Gather sources: Use WebSearch for external data (benchmarks, docs, comparisons); WebFetch for full page content from URLs; ReadArtifact for dependency outputs; ReadFile for local files. Extract key points from each.
  3. Synthesize: Combine findings. Note agreements, conflicts, and gaps. Use tables for side-by-side comparison.
  4. Cite: When referencing sources from input artifacts, include clear attribution. Required formats:
    • Inline: [Task N] or [Source Name] for artifact-derived claims
    • Quantitative claims (RPS, benchmarks, metrics): must have attribution—never state numbers without a source
    • End of report: ## References section listing all sources (artifact names, URLs if available, or "Official documentation" for general knowledge)

Synthesis Techniques

  • Thematic synthesis: Group findings by theme or topic. Preferred when many sources.
  • Comparative synthesis: Side-by-side table when comparing options (technologies, approaches).
  • Chronological: When timeline or evolution matters.
  • Gap analysis: What is known vs unknown; recommend next steps.
  • Pro-con synthesis: Aggregate pros/cons across sources when comparing alternatives.

Output Structure

Section Purpose
Findings Organized by topic or source. Use subsections. Use inline [Source] for attribution.
Summary Key takeaways in bullet or paragraph form.
Gaps What could not be determined; limitations.
Recommendations Next steps, conclusions, or suggested actions.
References Required. List all sources: [Task N] artifact, [Official Docs] framework name, or URL when available.

Handling Conflicting Sources

When sources disagree:

  • Present both views with attribution.
  • Note the conflict explicitly.
  • If possible, suggest resolution (e.g. "more recent", "broader sample").
  • Do not arbitrarily pick one; let the reader see the conflict.

Sandbox Usage

  • Save intermediate notes to sandbox/notes.md
  • Store raw or structured data to sandbox/data.json if needed
  • Draft sections in sandbox/draft.md before finalizing
  • Final output via Finish tool

Quality Checklist

  • Findings organized logically
  • All sources attributed—inline [Source] for claims; no unsourced quantitative data
  • ## References section present and non-empty
  • Summary captures main points
  • Gaps/limitations acknowledged when relevant
  • Recommendations actionable when appropriate
  • Conflicts between sources noted when present