Academic Research Agent Skill

A Claude-compatible academic research agent skill for PhD and Master students: literature review, novelty checks, research Reality Gates, feasibility testing, experiment planning, reviewer simulation, and claim verification.

Use agent research skills to read deeper, think wider, plan better, and review harder while the researcher stays in control.

Academic Research Agent Skill helps Master, PhD, and independent researchers in technical disciplines collaborate with AI agents across the research lifecycle: scoping, literature review, research paper planning, novelty checking, mathematical formalization, experiment planning, reviewer simulation, and claim verification. It is designed for Claude-style skills, Claude Code workflows, ChatGPT/Codex-style agents, Gemini, and local LLM research assistants.

It is not an autonomous paper factory or an “auto research paper” generator. The goal is not to remove the researcher. The goal is to give researchers a disciplined collaboration workflow so they can learn faster, explore more ideas, challenge weak assumptions earlier, and complete research with stronger evidence.

Why Students Use It

Who It Is For

• Master students learning how to structure a serious research project.
• PhD students managing literature, novelty, experiments, and writing pressure.
• Students in computer science, AI, machine learning, mathematics, engineering, and related technical fields.
• Research engineers turning rough ideas into executable experiments.
• Supervisors who want traceable AI-assisted research artifacts.
• Students and researchers who need a reusable research workflow.

What’s Included

Core Promise

flowchart LR
    H["Researcherjudgment, goals, approval"]  A["Agent teamreading, drafting, critique, planning"]
    A --> E["Evidence-traced artifacts"]
    E --> H
    H --> D["Better research decisions"]

The researcher provides direction, taste, constraints, and final judgment. The agents provide breadth, structure, critique, and execution support. Every important step creates an artifact that can be reviewed, revised, and learned from.

What This Skill Does

• Turns rough ideas into scoped research questions and contributions.
• Supports Claude skill for PhD student workflows, Claude skills for research paper planning, and agent research workflows.
• Forces source inspection before claims are used.
• Works with tool-assisted paper ingestion: download PDFs, convert them to Markdown, extract figures/tables, and build source analysis matrices.
• Grounds methods against closest prior work.
• Blocks shallow novelty with an explicit novelty gate.
• Requires the minimum mathematical definition needed to falsify a claim.
• Runs a Research Reality Gate before deep implementation planning.
• Separates technical smoke tests, scientific feasibility pilots, and full runs.
• Builds risk plans, work breakdowns, and code execution plans only for reality-cleared claims.
• Simulates reviewers before the paper is too expensive to fix.
• Verifies claims against sources or experiment artifacts.
• Supports configurable output language while keeping prompts stable in English.

Tool-Assisted Paper Ingestion

This skill is designed for real research workflows where papers are processed with tools, not only pasted into chat.

See references/tool_layer.md for the expected output contracts.

What This Skill Does Not Do

• It does not write a fake paper from thin air.
• It does not promise automatic research paper generation without human judgment, sources, methods, and verification.
• It does not invent citations, datasets, baselines, or results.
• It does not replace advisor feedback or human research judgment.
• It does not guarantee acceptance at any venue.
• It does not claim full autonomy.

Quick Start

Clone the repo into a research workspace used with Claude, ChatGPT, Gemini, local LLM agents, or another AI agent that can read repository instructions.

git clone 
cd Academic-Research-Agent-Skill
cp config/language.example.yaml config/language.yaml

Then ask your assistant:

Use SKILL.md and CLAUDE.md. Help me develop this research idea:
"My research topic here"

Recommended first sequence:

/paper-scope
/pdf-ingest
/lit-ground
/math-formalize
/astar-novelty
/reality-gate
# If authorized: experimental-unit audit and feasibility pilot
/risk-plan
/code-exec-plan
/reviewer-sim
/claim-verify

Language Configuration

Prompts are written in English for consistency. Outputs can be localized.

output_language: "Vietnamese"
secondary_language: "English"
translation_mode: "technical-terms-in-english"

See docs/LANGUAGE_CUSTOMIZATION.md.

Workflow

flowchart TD
    I["1. Ideahuman intent"] --> S["2. Scopequestion, contributions, non-goals"]
    S --> G1{"Human gateIs this worth pursuing?"}
    G1 -->|revise| S
    G1 -->|approve| L["3. Source ingestionpapers, notes, code, datasets"]
    L --> LG["4. Literature groundingclosest prior work and baselines"]
    LG --> M["5. Math formalizationdefinitions, objective, assumptions"]
    M --> N{"6. Novelty gatenot just method A on problem B"}
    N -->|fail| S
    N -->|conditional| F["Fix novelty or scope"]
    F --> N
    N -->|pass| G2{"7. Research Reality Gatewhat does evidence authorize?"}
    G2 -->|block| E["Bounded evidence recovery"]
    E --> G2
    G2 -->|pilot only| X["8. Claim-eligible feasibility pilot"]
    X --> G2
    G2 -->|execution ready| P["9. Claim freeze, risk and work plan"]
    P --> C["10. Approved code and experiment plan"]
    C --> G3{"Full-run gatePilot and provenance sufficient?"}
    G3 -->|no| X
    G3 -->|yes| R["11. Approved run and reviewer simulation"]
    R --> V["12. Claim verification"]
    V --> O["13. Draft, revise, submit, or archive"]

Agent Collaboration Model

graph TB
    H["Human Researcherdirection, constraints, approval"]
    O["Orchestratorstate, routing, gates"]
    S["Strategistscope, literature, math"]
    C["Criticnovelty, reviews, claim checks"]
    P["Plannerrisks, milestones, WBS"]
    A["Architectcode and experiment design"]
    E["Executorimplementation and pilots"]
    D["DevOpsenvironment and remote runs"]

    H  O
    O --> S
    O --> C
    O --> P
    O --> A
    O --> E
    O --> D
    S  C
    C --> H
    P --> H
    A --> H
    E --> H

Repository Map

.
├── SKILL.md                   # Core skill entry point
├── CLAUDE.md                  # Agent session instructions
├── references/                # Skill references loaded when needed
├── .claude/commands/          # Slash-command style prompts
├── _agents/                   # Role contracts, rules, workflows
├── config/                    # Language configuration
├── docs/                      # User-facing documentation
├── examples/                  # Topic brief and sample output structure
└── assets/image-prompts/      # Prompts for generating repo visuals

Why This Is Different

Expected Artifacts

• 02_Scope.md
• 05_Lit_Grounding.md
• 06_Math_Formalization.md
• 20_Reality_Gate.md
• 22_Experimental_Unit_Audit_Plan.md
• 23_Feasibility_Pilot_Protocol.md
• 10_Risk_Plan.md
• 11_WorkBreakdown.md
• 12_Code_Execution_Plan.md
• 14_Agent_Brief_PhaseN.md
• 15_Changelog.md
• 19_Source_Analysis_Matrix.md
• PDF-to-Markdown notes, figure/table reports, and download logs when tool-assisted ingestion is used
• Reviewer simulation and claim verification report

Visual Assets

The repository includes Mermaid diagrams and image-generation prompts:

• docs/visuals/human_agent_collaboration.mmd
• docs/visuals/research_lifecycle.mmd
• docs/visuals/artifact_pipeline.mmd
• assets/image-prompts/README.md

Documentation

• Getting Started
• Research Tools
• Positioning
• Use Cases
• Language Customization
• Visual Guide
• Image Generation Guide
• Competitive Analysis
• Launch Checklist

Recommended GitHub Description

Claude-compatible academic research agent skill for PhD/Master students: literature review, research paper planning, novelty checks, experiment planning, reviewer simulation, and claim verification.

Search Phrases This Repo Serves

This repository is relevant to searches such as Claude skill for PhD student, Claude skills for research paper, agent research skills, agent research workflow, academic research agent, AI research assistant for literature review, PhD research workflow, novelty check, claim verification, and auto research paper alternatives.

Safety Position

Academic Research Agent Skill is designed for evidence-traced collaboration. If a claim cannot be linked to an inspected source, formal artifact, or experiment result, it must be labeled as a hypothesis or removed.

Citation

If you use this repository in academic work, cite it as a human-guided academic research agent skill for Master and PhD students in technical fields.