agent-skills
AGENTS.md30 starsChanged 10 months agoMITView on GitHub
This repository is a curated collection of AI agent skills organized as a Claude Plugin marketplace. Each skill is a self-contained knowledge base combining: - Documentation (SKILL.md + references/) - Helper scripts (scripts/ directory) - Templates/boilerplate (assets/ directory) Refer to the README.md for installation instructions and usage tips. Skills are organized into 6 plugin bundles in .claude-plugin/marketplace.json: Each skill follows this structure: Key principle: SKILL.md is the starting point. References are for deep dives when needed. Each skill is completely…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository13 other files it gives its agents.
- cursor-agent-supervisorSkill
- github-pr-workflowSkill
- jj-todo-workflowSkill
- nix-profile-managerSkill
- nushell-plugin-builderSkill
- nushell-usageSkill
- package-npm-nixSkill
- read-bin-docsSkill
- research-assistantSkill
- searxng-searchSkill
- textual-builderSkill
- typst-writerSkill
- working-with-jjSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
26 rules. About 0 are common, 21 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈21
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 68
When creating or modifying skills:
- Distinctive: line 71
Start with frontmatter (name, description, version targets)
- Distinctive: line 78
Use clear section headings
- Very specific: line 83
Add executable permissions (chmod +x)
- Very specific: line 85
Use bash/python/nushell depending on complexity
- Distinctive: line 86
Keep scripts focused on single tasks
- Distinctive: line 91
Use placeholder names that are easy to replace
- Distinctive: line 97
Accuracy: Verify commands and examples work
- Distinctive: line 98
Completeness: Ensure key workflows are documented
- Distinctive: line 99
Clarity: Check that explanations are clear for agents
- Distinctive: line 100
Scripts: Test helper scripts with various inputs
- Very specific: line 104
When adding/removing skills, update .claude-plugin/marketplace.json:
- Distinctive: line 105
Add skills to appropriate plugin bundle
- Distinctive: line 107
Keep keywords relevant for discovery
- Distinctive: line 112
Markdown: All documentation (CommonMark/GFM)
- Distinctive: line 132
No test suite to run
- Distinctive: line 133
No compilation or bundling required
- Very specific: line 138
When creating SKILL.md files, use this frontmatter template:
- Distinctive: line 154
When the user asks about working with a technology:
- Distinctive: line 155
Check if a skill exists for it (see plugin list above)
- Very specific: line 156
Read the SKILL.md first
- Distinctive: line 157
Read references/ if more depth needed
- Distinctive: line 162
When you learn something new while working on tasks:
- Distinctive: line 164
If it's a complex workflow, suggest a new reference doc
- Distinctive: line 165
If it's a repetitive task, suggest a helper script
- Distinctive: line 166
If it's a common starting point, suggest a template
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
This file provides guidance to AI agents when working with code in this repository.
Project Overview
This repository is a curated collection of AI agent skills organized as a Claude Plugin marketplace. Each skill is a self-contained knowledge base combining:
- Documentation (SKILL.md + references/)
- Helper scripts (scripts/ directory)
- Templates/boilerplate (assets/ directory)
Refer to the README.md for installation instructions and usage tips.
Repository Structure
Plugin Organization
Skills are organized into 6 plugin bundles in .claude-plugin/marketplace.json:
- jujutsu-skills:
working-with-jj,jj-todo-workflow - nix-skills:
nix-profile-manager,package-npm-nix - python-skills:
textual-builder - nushell-skills:
nushell-plugin-builder - github-skills:
github-pr-workflow - typst-skills:
typst-writer
Skill Directory Pattern
Each skill follows this structure:
<skill-name>/
├── SKILL.md # Primary knowledge base (entry point)
├── scripts/ # [Optional] Helper utilities
│ └── <utility-scripts> # Bash/Python/Nushell scripts
├── assets/ # [Optional] Templates and boilerplate
│ └── <template-name>/ # Reusable project templates
└── references/ # [Optional] Deep-dive technical documentation
└── <topic>.md # Supplementary reference materialsKey principle: SKILL.md is the starting point. References are for deep dives when needed.
Architecture and Design Patterns
1. Self-Contained Skills
Each skill is completely independent and can be read/used without dependencies on other skills. Skills do not import or reference each other.
2. Layered Documentation
- SKILL.md: Quick reference, essential commands, common patterns
- references/: Detailed explanations, edge cases, advanced usage
- scripts/: Automation for repetitive tasks (not mandatory to understand the skill)
3. Version-Controlled Knowledge
This repository itself uses JJ (Jujutsu) version control colocated with Git. Changes to skills are tracked with granular commits showing:
- Documentation improvements
- New helper scripts
- Template updates
- Skill reorganizations
Working with Skills
Creating or Updating Skills
When creating or modifying skills:
- SKILL.md requirements:
- Start with frontmatter (name, description, version targets)
- Include quick reference tables for commands/syntax
- Provide examples of common workflows
- Link to reference docs for deep dives
- Reference documentation:
- Break complex topics into separate .md files
- Use clear section headings
- Include code examples with explanations
- Cross-reference related topics
- Helper scripts:
- Add executable permissions (
chmod +x) - Include usage comments at the top
- Use bash/python/nushell depending on complexity
- Keep scripts focused on single tasks
- Add executable permissions (
- Templates in assets/:
- Provide a complete working example
- Include a README.md explaining customization
- Use placeholder names that are easy to replace
- Document any dependencies
Testing Skills
Skills are primarily documentation, so "testing" means:
- Accuracy: Verify commands and examples work
- Completeness: Ensure key workflows are documented
- Clarity: Check that explanations are clear for agents
- Scripts: Test helper scripts with various inputs
Updating Plugin Metadata
When adding/removing skills, update .claude-plugin/marketplace.json:
- Add skills to appropriate plugin bundle
- Update version numbers
- Keep keywords relevant for discovery
Technology Stack
Languages
- Markdown: All documentation (CommonMark/GFM)
- Bash/Shell: Most helper scripts
- Python: Complex scripts (e.g., template generators)
- Rust: Nushell plugin template
- Nix: Package management integration
External Tools Referenced in Skills
- jj (Jujutsu v0.36.x): Version control
- gh: GitHub CLI
- nix: Package manager
- nushell: Shell with structured data
- cargo: Rust build system
- uv/uvx: Fast Python package installer
Important Notes
No Traditional Build Process
This is not a software project with tests or builds. It's a knowledge repository:
- Skills are documentation + reference implementations
- No test suite to run
- No compilation or bundling required
- Validation happens through practical use by agents
Skill Frontmatter Format
When creating SKILL.md files, use this frontmatter template:
---
name: skill-name
description: Brief description of what this skill provides
version: 1.0.0 # Skill version
targets:
- tool-name: ">=1.2.3" # Minimum versions of referenced tools
---Common Patterns When Working Here
Reading Skills
When the user asks about working with a technology:
- Check if a skill exists for it (see plugin list above)
- Read the SKILL.md first
- Read references/ if more depth needed
- Examine scripts/ to understand automation available
Improving Skills
When you learn something new while working on tasks:
- Propose adding it to the relevant skill
- If it's a complex workflow, suggest a new reference doc
- If it's a repetitive task, suggest a helper script
- If it's a common starting point, suggest a template
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

