scientific-thinking — Structured Scientific Reasoning for AI Agents

中文文档

What it does

• Frames and decomposes research questions before answering
• Distinguishes observed fact / direct evidence / indirect evidence / hypothesis / speculation
• Labels claim provenance: provided data vs. background knowledge vs. inference
• Ranks competing explanations by available support
• Calibrates conclusion language to evidence strength
• Defines interpretation boundaries and unresolved uncertainty
• Suggests the lowest-cost next step that would discriminate between explanations

Multi-Platform Support

Works with all major AI agents that support the Agent Skills format:

Platform	Status	Details
Claude Code	✅ Full support	Native SKILL.md format
OpenClaw / ClawHub	✅ Full support	metadata.openclaw namespace
Hermes Agent	✅ Full support	metadata.hermes namespace, category: research
Pi-Mo	✅ Full support	metadata.pimo namespace
OpenAI Codex	✅ Full support	agents/openai.yaml sidecar
SkillsMP	✅ Indexed	GitHub topics configured

Comparison: with vs. without this skill

Capability	Native agent	This skill
Distinguish fact from interpretation	Sometimes	Always, explicitly labeled
Label claim provenance	No	Yes — data / background / inference
Consider alternative explanations	Inconsistent	Yes — ranked by support
Calibrate language to evidence	No	Yes — 5-level scale
State interpretation boundary	Rarely	Always
Suggest discriminating next step	No	Yes
Handle null results correctly	No	Yes — absence ≠ evidence of absence
Reconcile conflicting papers	No	Yes — maps experimental differences
Evidence provenance for missing data	No	Yes — labels answer as provisional

When to use

• Interpreting experimental results or paper conclusions
• Evaluating competing hypotheses
• Analyzing mechanisms or pathways
• Designing or critiquing experiments
• Constructing scientific arguments for writing
• Reconciling conflicting findings
• Any research question where overclaiming is a risk

Skill Installation

Claude Code

# Global install (available in all projects)
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git ~/.claude/skills/scientific-thinking

# Project-level install
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git .claude/skills/scientific-thinking

OpenClaw / ClawHub

# Via ClawHub
clawhub install scientific-thinking

# Manual install
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git ~/.openclaw/skills/scientific-thinking

# Project-level install
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git skills/scientific-thinking

Hermes Agent

git clone https://github.com/Agents365-ai/scientific-thinking-skill.git ~/.hermes/skills/research/scientific-thinking

Or add to ~/.hermes/config.yaml:

skills:
  external_dirs:
    - ~/myskills/scientific-thinking

Pi-Mo

git clone https://github.com/Agents365-ai/scientific-thinking-skill.git ~/.pimo/skills/scientific-thinking

OpenAI Codex

# User-level install
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git ~/.agents/skills/scientific-thinking

# Project-level install
git clone https://github.com/Agents365-ai/scientific-thinking-skill.git .agents/skills/scientific-thinking

SkillsMP

skills install scientific-thinking

Installation paths summary

Platform	Global path	Project path
Claude Code	~/.claude/skills/scientific-thinking/	.claude/skills/scientific-thinking/
OpenClaw	~/.openclaw/skills/scientific-thinking/	skills/scientific-thinking/
Hermes Agent	~/.hermes/skills/research/scientific-thinking/	Via external_dirs config
Pi-Mo	~/.pimo/skills/scientific-thinking/	—
OpenAI Codex	~/.agents/skills/scientific-thinking/	.agents/skills/scientific-thinking/

Files

• SKILL.md — the only required file. Loaded by all platforms as the skill instructions.
• agents/openai.yaml — OpenAI Codex-specific configuration (display, policy, capabilities)
• checks.md — 10-point self-check list referenced by SKILL.md
• examples.md — 8 annotated examples referenced by SKILL.md
• README.md — this file (English)
• README_CN.md — Chinese documentation

Note: Only SKILL.md is needed for the skill to work. All other files are supplementary.

GitHub Topics

For SkillsMP indexing, this repository uses the following topics:

claude-code claude-code-skill claude-skills agent-skills skillsmp skill-md scientific-thinking research reasoning evidence-evaluation

License

MIT

Support

If this skill helps your research, consider supporting the author:

WeChat Pay

Alipay

Buy Me a Coffee

Author

Agents365-ai

• Bilibili: https://space.bilibili.com/441831884
• GitHub: https://github.com/Agents365-ai