Skip to main content

​
What It Does

Researcher Hand is a deep autonomous research agent that conducts exhaustive investigations, cross-references sources, fact-checks claims, and produces comprehensive structured reports with citations. Unlike a simple web search, Researcher Hand:
  • Decomposes complex questions into sub-questions
  • Gathers 5-50+ sources depending on depth setting
  • Evaluates source quality using CRAAP criteria (Currency, Relevance, Authority, Accuracy, Purpose)
  • Cross-references claims across multiple independent sources
  • Generates structured reports in multiple formats (brief, detailed, academic, executive)
  • Tracks its own research history in a knowledge graph

​
Activation

# Activate the Researcher Hand
openfang hand activate researcher

# Activate with custom settings
openfang hand activate researcher --settings "research_depth=exhaustive,output_style=academic"

# Check status
openfang hand status researcher

​
Configuration Settings

​
research_depth
select
default:"thorough"
How exhaustive each investigation should be:
  • quick: 5-10 sources, single pass
  • thorough: 20-30 sources, cross-referenced (default)
  • exhaustive: 50+ sources, multi-pass, fact-checked
​
output_style
select
default:"detailed"
How to format research reports:
  • brief: Executive summary, 1-2 pages
  • detailed: Structured report, 5-10 pages (default)
  • academic: Formal paper style with citations
  • executive: Key findings + recommendations
​
source_verification
toggle
default:"true"
Cross-check claims across multiple sources before including them in the report.
​
max_sources
select
default:"30"
Maximum number of sources to consult per investigation:
  • 10 sources
  • 30 sources (default)
  • 50 sources
  • Unlimited
​
auto_follow_up
toggle
default:"true"
Automatically research follow-up questions discovered during investigation.
​
save_research_log
toggle
default:"false"
Save detailed search queries and source evaluation notes to a log file.
​
citation_style
select
default:"inline_url"
How to cite sources in reports:
  • inline_url: Inline URLs (default)
  • footnotes: Footnote references
  • academic_apa: Academic APA format
  • numbered: Numbered references
​
language
select
default:"english"
Primary language for research and output:
  • English (default)
  • Spanish
  • French
  • German
  • Chinese
  • Japanese
  • Auto-detect

​
Required Tools

Researcher Hand requires access to these tools (all built-in):
  • shell_exec — Platform detection and system commands
  • file_read, file_write, file_list — Report generation
  • web_fetch, web_search — Source gathering
  • memory_store, memory_recall — State persistence
  • schedule_create, schedule_list, schedule_delete — Task scheduling
  • knowledge_add_entity, knowledge_add_relation, knowledge_query — Knowledge graph
  • event_publish — Completion notifications

​
System Prompt Overview

Researcher Hand operates in 7 phases:
1

Platform Detection & Context

Detects OS, loads cumulative research stats from memory, reads user configuration, queries knowledge graph for existing research on the topic.
2

Question Analysis & Decomposition

Identifies question type (factual, comparative, causal, predictive, how-to, survey), decomposes into 2-5 sub-questions, identifies authoritative source types for the topic.
3

Search Strategy Construction

Builds 3-5 search queries per sub-question using multiple strategies: direct, expert, comparison, temporal, and deep queries.
4

Information Gathering (Core Loop)

Executes searches, evaluates each source using CRAAP criteria (Currency, Relevance, Authority, Accuracy, Purpose), extracts key claims and data points. Continues until depth target is met.
5

Cross-Reference & Synthesis

Verifies claims appear in 2+ independent sources if verification is enabled. Groups findings by sub-question, identifies consensus and minority views, builds knowledge graph.
6

Fact-Check Pass

Searches for primary sources, checks for debunkings, verifies statistics, marks each claim with confidence level (Verified, Likely, Unverified, Disputed).
7

Report Generation

Generates report in configured format, saves to markdown file, updates dashboard statistics.

​
Usage Examples

​
Basic Research Query

openfang chat researcher
> "What are the key differences between Rust and Go for systems programming?"
Researcher Hand will:
  1. Decompose into sub-questions (performance, memory safety, concurrency models, ecosystem maturity)
  2. Search for benchmarks, engineering blogs, official docs, and expert comparisons
  3. Cross-reference claims across multiple sources
  4. Generate a detailed report with citations
  5. Save to research_rust_vs_go_YYYY-MM-DD.md

​
Academic-Style Report

openfang hand configure researcher --set output_style=academic --set citation_style=academic_apa
openfang chat researcher
> "What is the current state of research on AI alignment?"
Output format: 
# AI Alignment Research: Current State
## Abstract
## Introduction
## Methodology
## Findings
## Discussion
## Conclusion
## References (APA format)

​
Exhaustive Investigation

openfang hand configure researcher --set research_depth=exhaustive --set max_sources=50
openfang chat researcher
> "Evaluate the evidence for and against intermittent fasting for longevity."
Researcher Hand will:
  • Consult 50+ sources including academic papers, meta-analyses, clinical trials
  • Fact-check all claims against primary sources
  • Present both supporting and contradicting evidence
  • Note gaps in current research

​
Dashboard Metrics

Researcher Hand tracks four key metrics:

Queries Solved

Total number of research questions answered.

Sources Cited

Total unique sources cited across all reports.

Reports Generated

Total research reports produced.

Active Investigations

Currently in-progress research tasks.
View in the dashboard at http://localhost:4200/hands/researcher.

​
Source Quality Evaluation (CRAAP Framework)

Researcher Hand scores each source using the CRAAP test:
When was it published or last updated? Is the information still current for the topic?For technology topics, anything >2 years old may be outdated.
Does it directly address your question? Who is the intended audience? Is the level of detail appropriate?
Who is the author? What are their credentials? What institution published this? Does the URL domain indicate authority (.gov, .edu, reputable org)?
Is the information supported by evidence? Has it been reviewed or refereed? Can you verify the claims from other sources?
Why does this information exist? Is it informational, commercial, persuasive, or entertainment? Is the bias clear or hidden?
Sources are graded:
  • A (Authoritative): Passes all 5 criteria
  • B (Reliable): Passes 4/5, minor concern on one
  • C (Useful): Passes 3/5, use with caveats
  • D (Weak): Passes 2/5 or fewer
  • F (Unreliable): Do not cite

​
Report Formats

​
Brief Report

# Research: [Question]
## Key Findings
- [3-5 bullet points with the most important answers]
## Sources
[Top 5 sources with URLs]

​
Detailed Report (Default)

# Research Report: [Question]
**Date**: YYYY-MM-DD | **Sources Consulted**: N | **Confidence**: [high/medium/low]

## Executive Summary
[2-3 paragraphs synthesizing the answer]

## Detailed Findings
### [Sub-question 1]
[Findings with citations]

## Key Data Points
| Metric | Value | Source | Confidence |

## Contradictions & Open Questions
[Areas where sources disagree or gaps exist]

## Sources
[Full source list with quality ratings]

​
Academic Report

# [Title]
## Abstract
## Introduction
## Methodology
## Findings
## Discussion
## Conclusion
## References (APA format)

​
Executive Report

# [Question] — Executive Brief
## Bottom Line
[1-2 sentence answer]
## Key Findings (bullet points)
## Recommendations
## Risk Factors
## Sources

​
Best Practices

Researcher Hand will never fabricate sources or citations. Every claim is traceable. If no reliable sources are found, it will say so explicitly.
Enable save_research_log to see every query and source evaluation. Useful for understanding the research process or debugging.
For controversial topics, Researcher Hand presents multiple perspectives fairly. It distinguishes between facts, expert opinions, and its own analysis.

​
Advanced Configuration

​
Custom Citation Formats

Edit ~/.openfang/hands/researcher.toml to add custom citation formats: 
[[settings]]
key = "citation_style"
setting_type = "select"
default = "inline_url"

[[settings.options]]
value = "chicago"
label = "Chicago Manual of Style"

​
Multi-Language Research

For non-English research, set the language: 
openfang hand configure researcher --set language=spanish
Researcher Hand will search in the target language and produce reports in that language.

​
Example Output

Here’s a snippet from a detailed report: 
# Research Report: Rust vs Go for Systems Programming
**Date**: 2025-03-07 | **Sources Consulted**: 28 | **Confidence**: high

## Executive Summary
Rust and Go both target systems programming but make different tradeoffs. Rust prioritizes zero-cost abstractions and memory safety through its ownership system, achieving C/C++-level performance. Go prioritizes simplicity and fast compilation, using garbage collection and a simpler concurrency model. Rust is better suited for performance-critical systems (game engines, operating systems, embedded), while Go excels at networked services and microservices where fast iteration and ease of deployment matter more than peak performance.

## Detailed Findings

### Performance Characteristics
Rust achieves comparable performance to C/C++ (within 5-10% in most benchmarks[1][2]). Go is typically 2-3x slower than Rust in CPU-intensive tasks[3], primarily due to garbage collection overhead and lack of zero-cost abstractions.

Key benchmarks:
| Benchmark | Rust | Go | Source | Confidence |
|-----------|------|----|-

Build docs developers (and LLMs) love

Get started for freeTalk to us