skills.rest
0h-n0
0h-n0COMMUNITY
Explore all skills by this author →
GitHub

auto-research

Orchestrates LLM research from literature survey to paper draft via an 8-phase state machine.

Updated May 9, 2026
Search Tags:#workflow automation#experiment design#reproducibility#state machine#paper drafting#literature survey#llm research
One-click install
npx skills add https://github.com/0h-n0/auto-research --skill auto-research-0h-n0
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: auto-research
Source: https://github.com/0h-n0/auto-research/tree/main/skills/auto-research
Command: npx skills add https://github.com/0h-n0/auto-research --skill auto-research-0h-n0

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Conducting LLM research end-to-end—surveying papers, finding gaps, designing experiments, running ablations, and drafting a paper—involves many disconnected steps that are easy to lose track of. This Skill manages the full lifecycle as a structured state machine with persistent project state, so nothing is lost and every decision is recorded. ## Core Features & Use Cases - 8-Phase / 4-Gate Workflow: Progresses from topic framing through literature survey, gap analysis, experiment design, TDD-based implementation, run analysis, paper drafting, and self-review, pausing at four user approval gates. - Persistent Project State: Tracks progress in .research/<slug>/STATE.json with rollback edges, so interrupted projects resume exactly where they stopped. - Reproducibility & Lab Notebook: Enforces seed/config/git-rev pinning, generates postmortems for failed runs, and maintains a decision journal with predicted-vs-actual metacognition. - Use Case: Start a new research project from an arXiv URL or topic string, let the workflow survey related work via arxiv-mcp-agent, propose scored ideas, design a budget-constrained experiment, and produce a NeurIPS/ACL-formatted paper draft with a self-review report. ## Quick Start Start a new LLM research project on the topic "attention sinks in long-context Llama models" and run Phase 1 topic framing.

Frequently Asked Questions about auto-research

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I start an automated LLM research workflow from a paper or topic?▼

Invoke the skill with a natural-language research theme, an arXiv URL, or --resume <slug> for an existing project. It generates a project slug, initializes .research/<slug>/ with STATE.json, drafts a research brief, and asks for scope confirmation at Gate G1.

How does the workflow handle literature surveys and gap analysis?▼

Phase 2 dispatches arxiv-mcp-agent for breadth-first search and paper-deep-reader agents in parallel for deep reading, producing a comparison matrix. Phase 3 runs three parallel research-gap-finder agents with different seeds to propose scored, falsifiable ideas.

Can I resume an interrupted research project?▼

Yes. Pass --resume <slug> and the skill reads .research/<slug>/STATE.json, then restarts from the phase indicated by current_phase and last_gate_passed. Completed projects with completed_at set are reported as finished.

What happens when an experiment run fails or sanity checks fail?▼

Failed runs are kept with STATUS=failed, a POSTMORTEM.md draft, and a reproduce.sh script. Sanity check failures trigger a rollback from Phase 6 to Phase 5, recorded in STATE.json rollbacks and CHANGELOG.md.

What are the prerequisites and limitations of this research workflow?▼

It requires Python with uv, PyTorch, HuggingFace Transformers, and a pre-configured arxiv-mcp-server. It is not intended for one-off paper summaries, production ML inference pipelines, or non-LLM research domains.

Related Skills

Explore More Education & Research Skills→
NeoLabHQNeoLabHQ
1.5k

setup-arxiv-mcp

Configures the paper-search MCP server via Docker MCP for academic paper search.

Official
Intermediate
fundamental-physicsfundamental-physics
17

zenodo

Query Zenodo records and download datasets, software, or publications.

Official
Intermediate
JesusFigueroa25JesusFigueroa25

mlops-engineer

Automate ML infrastructure setup and MLOps workflows across AWS, Azure, and Google Cloud.

Community
Advanced
raja21068raja21068
2

alphaxiv

Summarize an arXiv research paper from its ID or URL.

Community
Intermediate