Why ARG?
AI is your copilot, not the pilot. ARG doesn't write your paper for you. It handles the grunt work — hunting down references, formatting citations, verifying data, checking logical consistency — so you can focus on what actually requires your brain: defining the question, choosing the method, interpreting the data, and writing the sentence after "I argue that..."
Fully autonomous AI research pipelines inherit structural failure modes: hallucinated results, methodology fabrication, frame-lock, citation hallucinations. These are not bugs — they're architectural.
ARG is built on one premise: a human researcher + AI avoids these failure modes better than either alone.
| Problem | ARG's Solution |
|---|
| Citation hallucinations | Semantic Scholar API verification + integrity gates |
| Frame-lock | Devil's Advocate with concession threshold protocol |
| Sycophancy | Anti-sycophancy scoring (1–5 scale before any concession) |
| Premature convergence | Intent detection + dialogue health monitoring |
| Methodology fabrication | 7-mode blocking checklist at Stage 2.5 & 4.5 |
| Unverifiable claims | Three-layer citation protocol + claim-faithfulness anchoring |
Quick Start
git clone https://github.com/Masqiller/ARG-RESEARCHER-V4.1.git ~/ARG-Researcher
Open the repo in your IDE. Config files are auto-discovered:
| Agent | Config file | How commands work |
|---|
| Google Antigravity | .agents/rules/arg-instructions.md + .agents/workflows/arg-*.md | Native /arg-* slash commands in Agent Manager |
| Windsurf | .windsurf/workflows/arg-*.md | Native /arg-* slash commands in chat |
| Cursor | .cursor/rules/arg-instructions.mdc | Type /arg-* in chat |
| Claude Code | .claude/CLAUDE.md + commands/ | Native /arg-* commands |
| GitHub Copilot | .github/copilot-instructions.md | Type /arg-* in chat |
| Cline / Roo Code | .clinerules | Type /arg-* in chat |
| Aider | .aider/conventions.md | Type /arg-* in chat |
| Continue | .continue/config.json | Type /arg-* in chat |
| Any other agent | .ai/instructions.md | Type /arg-* in chat |
Test it: Type /arg-plan and describe any paper topic. ARG will start a Socratic dialogue to map out your chapter structure in minutes.
Optional dependencies: Pandoc (DOCX export), tectonic (PDF/LaTeX). Markdown output works without either.
Detailed setup: UNIVERSAL_SETUP.md
Slash Commands
All commands follow the pattern /arg-<mode>. You can use them explicitly or just describe what you need in plain language — ARG detects your intent automatically.
/arg-full
Full end-to-end pipeline: research → write → review → revise → finalize
The complete 10-stage orchestrated workflow. Invokes all 4 skills in sequence.
/arg-full
