Idea Discovery Pipeline
Chains literature survey, idea brainstorming, novelty checks and review into one research workflow
- Stars
- 18,482
- Licence
- MIT
- Files
- 1 file
What it does
When installed, the agent runs a multi-step research workflow that starts from a broad research direction and ends with ranked, reviewed research ideas and a refined proposal and experiment plan for the top idea. It chains sub-skills for literature survey, brainstorming, novelty verification, external critical review and method refinement, saving outputs into an idea-stage folder with a run record. It can also run small pilot experiments on GPUs within time and budget limits, and can optionally render the final report to HTML.
Great for
- Researchers or PhD students starting a new project who want a structured path from topic to shortlisted ideas
- ML teams who want an automated first pass on a research direction before committing GPU time
Skip it if
- Users who need a fully hands-off result without checking it: the workflow depends on other sub-skills and an external reviewer model, and its output still needs expert judgement
- Anyone unwilling to run pilot experiments on their own compute, since pilots consume GPU hours and may take up to hours each
Try asking
- “Run the full idea discovery pipeline on: efficient fine-tuning for small language models”
- “Find and validate research ideas from scratch on: graph neural networks for molecular property prediction, with a reference paper at arxiv.org/abs/2406.04329”
- “Start idea discovery with compact summaries enabled for a short-context session on: multimodal retrieval”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
idea-discovery.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip idea-discovery.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip idea-discovery.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Set AUTO_PROCEED to false if you want to approve each selection checkpoint instead of letting the workflow continue automatically
- Keep ARXIV_DOWNLOAD false unless you need full PDFs, to save time and storage
- Review the pilot budget limits (2 hours per pilot, 8 GPU hours total) and adjust them to your compute before starting
What's inside
- SKILL.md
Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.