Agent skill
ai-research
Systematic AI research process. Use when the user is AI researcher and the project is involved in AI research project.
Install this agent skill to your Project
npx add-skill https://github.com/ParkJumyung/skills/tree/main/ai-research
SKILL.md
AI Research
A modular skill distilled from Neel Nanda’s research process sequence (Posts 1–3).
The goal is to help you make progress under uncertainty by identifying the right stage, the right north star, and the right next actions.
This skill is designed for empirical research with short feedback loops, but generalizes broadly.
North star
Research progress looks different at different stages:
- Explore: maximize information gained per unit time
- Understand: gather evidence that convinces you a hypothesis is true or false
- Distill: compress findings into concise, defensible, communicable truth
A common failure mode is using the wrong standard for the current stage.
How to use
Common inputs
- A short description of the project
- Current artifacts (results, plots, notes, draft text)
- What feels confusing or stuck
- Any constraints (deadline, compute, scope)
(Use: Intake template)
Common outputs
- Current stage + north star
- Top uncertainties
- 1–3 prioritized next actions
- A fast-fail test (if the direction is wrong)
- Baselines / alternative explanations to check
- If applicable: hypothesis or claim distillation
(Default output format: Triage output template)
Workflows
A) “I’m stuck / what should I do next?”
- Stage classifier
- Run the relevant stage module:
- Explore → Exploration: information gain
- Understand → Understanding: hypotheses & evidence
- Distill → Distillation: communicable truth
- Prioritisation
- Produce triage output
B) “I don’t trust my results”
- Truth-seeking
- Understanding: hypotheses & evidence
- Red-team checklist
C) “I need to move faster”
- Moving fast
- Tight feedback loops
- Fail fast
D) “I need to write this up / distill”
- Distillation: communicable truth
- Distillation claims template
- Hypothesis ↔ evidence map
Modules index
- 00 Overview
- 01 Stage classifier
- 02 Ideation & problem choice
- 03 Exploration: information gain
- 04 Understanding: hypotheses & evidence
- 05 Distillation: communicable truth
- 06 Truth-seeking
- 07 Prioritisation
- 08 Moving fast
- 09 Tight feedback loops
- 10 Fail fast
- 11 Action under uncertainty
- 12 Research taste: overview
- 13 Research taste: ingredients
- 14 Cultivating research taste
- 15 Common failure modes
Templates
- Intake template
- Triage output template
- Experiment plan template
- Red-team template
- Distillation claims template
- Weekly review template
Acknowledgements
Compiled from Neel Nanda’s research process sequence (Posts 1–3, Apr–May 2025).
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
ParkJumyung/skills
ai-paper-writing
Best practices for writing AI research papers. Use when the project involves writing a research paper in AI field.
mattpocock/skills
edit-article
Edit and improve articles by restructuring sections, improving clarity, and tightening prose. Use when user wants to edit, revise, or improve an article draft.
mattpocock/skills
setup-pre-commit
Set up Husky pre-commit hooks with lint-staged (Prettier), type checking, and tests in the current repo. Use when user wants to add pre-commit hooks, set up Husky, configure lint-staged, or add commit-time formatting/typechecking/testing.
mattpocock/skills
obsidian-vault
Search, create, and manage notes in the Obsidian vault with wikilinks and index notes. Use when user wants to find, create, or organize notes in Obsidian.
mattpocock/skills
scaffold-exercises
Create exercise directory structures with sections, problems, solutions, and explainers that pass linting. Use when user wants to scaffold exercises, create exercise stubs, or set up a new course section.
mattpocock/skills
git-guardrails-claude-code
Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, branch -D, etc.) before they execute. Use when user wants to prevent destructive git operations, add git safety hooks, or block git push/reset in Claude Code.
Didn't find tool you were looking for?