Modes for the RESEARCH Stage in the ARS Pipeline: The 7 Deep-Research Modes Explained

The RESEARCH stage in the Academic Research Skills (ARS) pipeline supports seven distinct modes—full, quick, review, lit-review, fact-check, socratic, and systematic-review—each designed for specific output formats ranging from 500-word briefs to 15,000-word PRISMA-compliant systematic reviews.

The Academic Research Skills (ARS) pipeline, maintained in the Imbad0202/academic-research-skills repository, structures complex research tasks into discrete stages. The RESEARCH stage is powered by the deep-research skill, which defines its operational behavior through a configurable mode system documented in MODE_REGISTRY.md.

Overview of the RESEARCH Stage Architecture

According to the source code in MODE_REGISTRY.md, the RESEARCH stage delegates execution to the deep-research skill bundle, which contains specialized agents such as research_question_agent and synthesis_agent. When you invoke the pipeline with --stage research, the --mode parameter determines which agent configuration activates, controlling the fidelity spectrum, oversight level, and output format. Detailed specifications for individual modes reside in commands/ars-full.md and commands/ars-lit-review.md, while deep-research/SKILL.md defines agent behaviors. The test suite in tests/fixtures/ validates each mode's output against these specifications.

The Seven RESEARCH Stage Modes

The deep-research skill registers seven distinct modes in the ARS pipeline. Each mode targets a specific research use case, output length, and level of human oversight.

full

Spectrum: Balanced
Typical Output: APA 7.0 research report, 3,000–8,000 words
Oversight: High
Trigger Phrases: "research [topic]", "deep research", "academic analysis"

The full mode generates comprehensive academic papers with proper citation formatting. It operates with high oversight, prompting for user confirmation at key decision points during the research and synthesis phases.

quick

Spectrum: Fidelity
Typical Output: Research brief, 500–1,500 words
Oversight: Medium
Trigger Phrases: "quick brief", "30 minute summary", "quick research"

Use quick mode for rapid intelligence gathering. This mode prioritizes speed and operates more autonomously than full, delivering condensed findings without extensive intermediate review steps.

review

Spectrum: Balanced
Typical Output: Peer reviewer-style report on provided text
Oversight: High
Trigger Phrases: "review this paper", "evaluate this paper", "assess this source"

The review mode transforms the pipeline into a critical assessment tool. It analyzes existing manuscripts or sources and produces structured reviewer reports, maintaining high oversight to ensure analytical rigor.

lit-review

Spectrum: Fidelity
Typical Output: Annotated bibliography plus synthesis narrative
Oversight: Medium
Trigger Phrases: "literature review", "annotated bibliography"

lit-review mode specializes in survey research. It aggregates sources into annotated bibliographies followed by integrative synthesis, operating with medium oversight to balance thoroughness with efficiency.

fact-check

Spectrum: Fidelity
Typical Output: Claim-by-claim verification report
Oversight: Medium
Trigger Phrases: "verify claims", "fact-check", "evidence verification"

The fact-check mode performs granular verification of factual assertions. It examines individual claims against evidence sources, producing detailed verification reports with medium autonomy.

socratic

Spectrum: Originality
Typical Output: Research-plan summary plus INSIGHT collection
Oversight: Very High
Trigger Phrases: "guide my research", "help me think through", "I'm not sure what to research"

socratic mode functions as an interactive research consultant. With very high oversight, it engages in dialogue to help shape research questions and methodologies, collecting insights rather than producing final documents.

systematic-review

Spectrum: Fidelity
Typical Output: PRISMA 2020 compliant report, 5,000–15,000 words
Oversight: Medium
Trigger Phrases: "systematic review", "meta-analysis", "PRISMA"

The systematic-review mode implements PRISMA 2020 guidelines for evidence synthesis. It generates comprehensive systematic reviews suitable for academic publication, operating with medium oversight to manage the complexity of protocol-driven research.

Invoking RESEARCH Stage Modes

You select modes via the ARS CLI using the --stage research --mode <mode> syntax. As implemented in the main branch, the following commands demonstrate each mode:

# Full research: Balanced depth with extensive APA formatting

ars pipeline --stage research --mode full

# Quick research: Concise 500-1,500 word brief

ars pipeline --stage research --mode quick

# Review mode: Peer-review assessment of provided text

ars pipeline --stage research --mode review

# Literature review: Annotated bibliography with synthesis

ars pipeline --stage research --mode lit-review

# Fact-check mode: Granular claim verification

ars pipeline --stage research --mode fact-check

# Socratic mode: Interactive research planning

ars pipeline --stage research --mode socratic

# Systematic review: PRISMA-compliant comprehensive analysis

ars pipeline --stage research --mode systematic-review

Mode Configuration and Pipeline Behavior

The mode selection directly influences which agents from the deep-research skill bundle activate. Higher-oversight modes (full, review, socratic) trigger interactive checkpoints where the pipeline pauses for user confirmation before proceeding to synthesis or conclusion phases. Lower-oversight modes (quick, lit-review, fact-check, systematic-review) run autonomously, generating finished artifacts with minimal human intervention.

Summary

• The RESEARCH stage in the ARS pipeline offers seven specialized modes: full, quick, review, lit-review, fact-check, socratic, and systematic-review.
• Mode definitions reside in MODE_REGISTRY.md, which serves as the single source of truth for the deep-research skill configuration.
• Oversight levels range from Medium (autonomous execution) to Very High (interactive dialogue), depending on the mode's fidelity and originality requirements.
• CLI invocation follows the pattern: ars pipeline --stage research --mode <mode_name>.

Frequently Asked Questions

What is the difference between high and medium oversight in RESEARCH modes?

High oversight modes (full, review, socratic) pause execution at critical decision points to request user confirmation before proceeding with synthesis or analysis steps. Medium oversight modes (quick, lit-review, fact-check, systematic-review) execute autonomously once triggered, delivering completed artifacts without intermediate user interaction.

How does the socratic mode differ from other RESEARCH modes?

Unlike output-oriented modes that produce final documents, socratic mode operates as an interactive research consultant. It generates research-plan summaries and collects INSIGHT objects through dialogue, utilizing very high oversight to guide users who have not yet defined specific research questions.

Where are the RESEARCH stage modes defined in the codebase?

The authoritative definitions for all seven modes are located in MODE_REGISTRY.md at the repository root. Detailed agent specifications for the deep-research skill that implements these modes are documented in deep-research/SKILL.md, with command-specific documentation available in commands/ars-full.md and commands/ars-lit-review.md.

Can I switch modes mid-pipeline during a RESEARCH stage execution?

The ARS pipeline architecture binds the mode parameter at invocation via --mode. To switch operational characteristics mid-process, you must terminate the current execution and re-invoke the pipeline with the desired mode flag, as the mode selection determines which agent configurations load at initialization.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}