Research Craft Skill

Research Craft is an agent skill for turning vague research work into a tight loop: choose better problems, improve inputs, forecast before experiments, keep logs, inspect failures, and publish useful artifacts.

It is based on Vivek’s article “how to be good at research”.

Install

Install everywhere the skills CLI can target:

npx -y skills add nik1t7n/research-craft-skill --all

Install only for common coding agents:

npx -y skills add nik1t7n/research-craft-skill \
  -a codex \
  -a claude-code \
  -a gemini-cli \
  -a antigravity \
  -a opencode \
  -a cursor \
  -y

Install for one agent:

npx -y skills add nik1t7n/research-craft-skill -a codex -y

The repository is a root skill bundle, so no subpath is required.

Use

Ask your agent to use research-craft when you are:

• choosing a research direction
• designing ML/AI experiments
• reviewing a research plan
• reading papers and building a source list
• tightening an experiment loop
• creating a research log
• analyzing failures or model outputs

Structure

.
├── SKILL.md
└── agents/openai.yaml

No scripts or assets are included because this skill is a compact workflow, not an automation package.

Compatibility

The skill uses the open SKILL.md format and is installable through the skills CLI.

Tested targets:

• Codex: codex
• Claude Code: claude-code
• Gemini CLI: gemini-cli
• Antigravity: antigravity
• OpenCode: opencode
• Cursor: cursor
• universal agent directory: .agents/skills

License

MIT