| name | workshop |
| description | Interactively guide the user to create a workshop/topic description file for the AI Scientist pipeline. Asks structured questions and generates a ready-to-use .md file. |
Workshop Description Creator
You are helping the user create a workshop/topic description file that will drive the AI Scientist research pipeline. This file defines the research scope — the AI Scientist will generate ideas, run experiments, and write papers within this scope.
Arguments
--output <path>: Output file path (default: examples/ideas/<slugified_title>.md)
Parse from the user's message. If no output is specified, derive the filename from the title.
Procedure
1. Ask the User
Guide the user through these questions one at a time (don't dump all questions at once). Adapt based on their answers — skip questions they've already addressed, probe deeper where needed.
a. Research Area
What research area or topic do you want to explore?
Examples: "failure modes of LLMs in code generation", "efficient training of vision transformers", "reinforcement learning for robotics manipulation"
b. Motivation
Why is this important? What problem are you trying to solve?
Help them articulate the gap or opportunity.
c. Scope
What specific aspects are you most interested in?
Narrow down from a broad area to actionable directions. For example, if they said "LLM safety", ask whether they mean alignment, jailbreaking, hallucination, or something else.
d. Desired Contributions
What kind of results would be valuable? (e.g., new methods, negative results, benchmarks, theoretical analysis)
e. Keywords
Can you list 3-5 keywords for this topic?
If the user struggles, suggest keywords based on what they've said.
2. Generate the Workshop Description
Using the user's answers, generate a .md file with this exact structure:
# Title: <Descriptive Title>
## Keywords
<keyword1>, <keyword2>, <keyword3>
## TL;DR
<One sentence summary>
## Abstract
<1-2 paragraphs: research context, motivation, open problems, what contributions are welcome>
Writing guidelines:
- Title: Specific and descriptive, not generic
- TL;DR: One sentence, accessible to non-experts
- Abstract: 150-300 words. Include: (1) why this area matters, (2) what open problems exist, (3) what kinds of ideas/experiments are in scope. Write in academic style but keep it readable.
3. Review with User
Show the generated description to the user. Ask:
Does this capture your research direction? Anything to adjust?
Iterate until they're satisfied.
4. Save the File
Write the final description to the output path:
mkdir -p $(dirname <output_path>)
Then write the file.
5. Next Steps
After saving, suggest:
Your workshop description is ready. You can now run the full pipeline:
claude "/ai-scientist --workshop <output_path>"
Or generate ideas first:
claude "/ai-scientist:ideation --workshop <output_path>"
Tips for Good Workshop Descriptions
If the user seems unsure, share these tips:
- Be specific about the domain — "deep learning for medical imaging" beats "AI applications"
- Mention open problems or surprising failures — these become seeds for novel ideas
- Include concrete examples of relevant topics or past work
- State what a good contribution looks like
- A template is available at
examples/workshop_template.md