Academic Research Agent Skill
A skill for master and PhD students that runs research through novelty and reality gates, from review to claim checks
- 71 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill reads and writes research notes and artifacts in the workspace folder
- When ingesting papers the agent visits arXiv and DOI links and runs conversion tools
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/academic-research-agent-skillDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads and writes research notes and artifacts in the workspace folder
- When ingesting papers the agent visits arXiv and DOI links and runs conversion tools
This is third-party code. Review the repository files before installing.
What it does
Academic Research Agent Skill turns an AI agent into a partner on a research project for master and PhD students and independent researchers in technical fields: computer science, AI, mathematics and engineering. It does not write a paper from thin air and promises no automatic result: decisions, goals and judgment stay with the researcher. The process runs through commands: `/paper-scope` frames the question, contribution and non-goals, `/pdf-ingest` processes papers, `/lit-ground` checks the method against closest work and baselines, `/math-formalize` gives the minimal definitions, `/astar-novelty` checks novelty and `/reality-gate` decides what the evidence authorizes: block, pilot only or execution ready. Then come `/risk-plan`, `/code-exec-plan`, `/reviewer-sim` and `/claim-verify`. `.claude/commands` holds 18 commands in all. With real papers the agent uses tools: it downloads PDFs from lists, arXiv links and DOI pages, converts them to Markdown, extracts figures and tables and builds a source analysis matrix. Output language is set in `config/language.yaml`, while prompts stay in English for stability.
Who it is for. Master and PhD students in computer science, AI and mathematics who want to go from a rough idea to verified claims without losing control of conclusions.
Good fit when
- The thesis topic is vague and you need a question, a contribution and a list of what the study does not do
- Before implementation you must learn whether the method amounts to applying a known trick to a new problem
- Twenty papers sit in PDF and the agent must turn them into notes with tables and figures
- Before deep planning you need a `/reality-gate` verdict: block, pilot or execution
- Before submission you need simulated reviewers and a check of claims against sources
Not a fit when
- You need automatic paper generation with no researcher involved: the author explicitly rejects that
- You work in the humanities: the skill targets technical disciplines
- You need a guarantee of acceptance at a venue: the project gives none
- The task is a one-off such as translating a paragraph: the full gated cycle is overkill
Example request
Use SKILL.md and CLAUDE.md and help me develop this idea: method X for task Y, start with /paper-scopeLimitations
- Clone the repository into the research workspace and copy `config/language.example.yaml` to `config/language.yaml`; the commands in `.claude/commands` target Claude Code and are pasted as prompts for other agents.
- PDF download, Markdown conversion and table extraction need separate tools from `references/tool_layer.md`; without them the agent works with what you paste.
- The skill does not invent citations, datasets or results, so when sources are missing it flags gaps and asks for material.
- The researcher always verifies conclusions: the project stresses that there is no full autonomy.
How to disable. Delete the repository folder from the workspace or remove the commands from `.claude/commands`. The skill installs nothing on the system. `config/language.yaml` and the research artifacts stay with you and are deleted by hand.
FAQ
How do I start with the skill?
Clone the repository, copy `config/language.example.yaml` to `config/language.yaml` and ask the agent to use `SKILL.md` and `CLAUDE.md` for your idea. The first command is `/paper-scope`.
What is the Reality Gate?
A gate before deep planning that tests the unit, metric, intervention, data access and valid yield. Verdicts: block, pilot only or execution ready.
Can I get output in Russian?
Yes. Set `output_language` and optionally `secondary_language` in `config/language.yaml` and keep `translation_mode` with English technical terms. Prompts stay in English.
Does the skill write the paper for me?
No. The project does not write a paper from nothing, invent citations, data or results, or replace your advisor.
How many commands are there?
Eighteen, in `.claude/commands`: from `/paper-scope` and `/pdf-ingest` to `/reviewer-sim`, `/claim-verify`, `/orchestrate` and `/remote-exec`.
Discussion
No questions or reviews yet.
Related
SkillsA library of 280+ research skills: literature search, experiment design, analysis, reference checks and paper preparation
A marketplace of five Claude Code plugins for researchers: literature search, research design, project memory and manuscript work
A Claude Code skill for scientific literature reviews: an interview, Scopus and OpenAlex search, DOI checks and a final docx file
Four skills for research text: drafting a paper from data, English polishing, removing templated phrasing and slides from a paper
paper-reading-skill turns one arXiv AI or robotics paper into a report with each claim tied to evidence in the source.
A skill set for academic work: paper review, pre-analysis plan check, grant proposal review, code audit and slide decks.
README in short
The README explains why students use such a skill and lists the contents: SKILL.md, CLAUDE.md, 18 commands, references, agent roles, examples and a project page. The principle of researcher control, PDF ingestion, quick start, language setup and a gated workflow diagram follow.
README badge
Original SKILL.md
--- name: research-agent description: Use this skill as a human-guided research agent for Master and PhD work in computer science, AI, mathematics, engineering, and related technical fields. Use it for research scoping, source ingestion, literature grounding, novelty review, mathematical formalization, reality and feasibility audits, experiment planning or execution, reframing a paper, reviewer simulation, and claim verification—especially when a plan looks polished but may depend on unverified data, metrics, interventions, access, resources, or compound hypotheses. --- # Research Agent Skill Move a technical research idea toward evidence-traced, falsifiable work while keeping research judgment with the human. ## Core contract - Let the researcher own direction, constraints, approvals, and final claims. - Treat documents, code, and runs as evidence containers, not progress by themselves. - Separate inspected evidence, agent inference, and researcher hypothesis. - Prefer the cheapest decisive falsifier before broad planning or implementation. - Stop, drop, or narrow work when a mandatory claim node fails; do not salvage every null result into a new story. - Maintain a traceable chain from research question and contribution to formal claim, closest competitor, falsifier or experiment/result ID, evidence artifact, paper location, and residual risk. ## Start every task 1. Read the project state and local instructions. 2. Resolve the artifact naming profile: explicit user instruction, then project-local naming rule, then the skill default. 3. Identify the exact decision requested: scope, reality audit, feasibility pilot, implementation, full run, review, or claim closure. 4. Read the smallest authoritative evidence set required for that decision. 5. State the current gate, allowed work, prohibited work, and next decisive action. Read [workflow.md](references/workflow.md) for lifecycle decisions. Read [reality_gate.md](references/reality_gate.md) before reframing, implementation planning, or any feasibility-sensitive experiment. Read the project-local file naming rule before creating or renaming artifacts. ## Default workflow 1. Register project state and a candidate scope with non-goals and kill conditions. 2. Ingest and inspect sources before using them as evidence. 3. Ground the candidate against closest prior work. 4. Formalize the minimum claim, measurement, and assumptions needed to falsify it.