Initialize a new research project by creating the project/ config directory. Run this ONCE at the very start of any new paper or research project — even if you have only a rough idea of the topic. All other skills read from project/ as the project grows; this skill just gets you started. Trigger when the user says any of the following: - "start new project" - "initialize project" - "set up project context" - "I'm starting a new paper" - "create project config" - "init project" - "new paper setup" - "set up my research project" - "configure project" - "project setup" - "beginning a new research project"
You are setting up the project/ config directory that all other research skills
depend on. This skill is intentionally lightweight — it only captures what you
actually know at day 1 of a research project.
The project/ directory is a living config. It starts sparse and fills in as
you make progress. Other skills (paper-search-and-triage, experiment-designer,
write-methodology-section, etc.) will add and update files here as your project
evolves. You do not need to know your system design, related work, baselines, or
results to run this skill.
Run this skill once per project. If project/research-focus.md already exists,
go to the Update flow at the bottom of this file.
Glob("project/*.md")
If project/research-focus.md exists, read it and say:
"I found an existing project config for [project_name]. Do you want to:
If the directory is empty or does not exist, proceed to Step 1.
Tell the user upfront:
"I'll ask you 3 short groups of questions to get your project started. You don't need to know everything yet — use 'TBD' or 'not sure yet' freely. These files will fill in as your project evolves."
project/research-focus.mdAsk:
Group 1 of 3 — Research Identity
- What do you want to call this project for now? (anything works — "paper1", "my-llm-project", a real system name if you have one)
- What general research area is this in? (e.g., "NLP", "systems security", "computer vision", "HCI")
- In 1–3 sentences: what problem are you trying to solve, or what question are you trying to answer? (rough is fine — even a direction counts)
- Do you have an initial idea for your approach or method, even vaguely? (say "not yet" if you're still exploring)
These four questions are the minimum. Everything else — system name, novelty, contributions, baselines — comes later through other skills.
project/venue-config.mdAsk:
Group 2 of 3 — Target Venue (optional)
- Do you have a target venue in mind? (e.g., ACL, NeurIPS, USENIX Security, ICSE — or say "unknown")
- Do you know the submission deadline? (YYYY-MM-DD, or "unknown")
- Is the submission double-blind (anonymous review)? (yes / no / unknown)
All answers can be "unknown". This file is easy to update later by saying "update venue config".
project/paper-paths.mdAsk:
Group 3 of 3 — Paper Files (optional)
Do you already have a LaTeX project set up? If yes, provide:
- Path to your main .tex file
- Path to your .bib file
- Path to your sections/ directory
- Path to your figures/ directory
If you haven't set up LaTeX yet, just say "not yet" and we'll leave these as TODO.
After collecting all answers, create the project/ directory and write 3 active
config files plus 5 placeholder files for configs that other skills will populate.
mkdir -p project
project/research-focus.md# Research Focus
project_name: {{Q1.1 answer}}
research_area: {{Q1.2 answer}}
last_updated: {{TODAY}}
## Problem Statement
{{Q1.3 answer — the problem or research question, in the student's own words}}
## Initial Approach
{{Q1.4 answer, or "Not yet defined — see paper-search-and-triage to explore the space"}}
## System Name
{{Q1.1 answer if a proper name, otherwise "TBD — will be defined as the project matures"}}
## What Makes This Novel
TBD — run paper-search-and-triage and research-gap-mapper to identify the gap
this work fills.
## Headline Result
TBD — fill in after experiments are complete.
project/venue-config.md# Venue Configuration