Skills

High-quality agent skills for research and engineering.

Skills encode the workflows, quality gates, and best practices senior engineers use, packaged so AI agents follow them consistently.

Why Skills?

AI coding agents default to the shortest path — skipping specs, tests, and reviews. Skills give agents structured workflows with steps, checkpoints, and verification so output stays production-grade.

Each skill is process, not prose: when to act, what to check, and what evidence is required before done.

Quick Start

Use with any agent that accepts Markdown instruction files.

Install with `npx`

npx skills add AIAI-Laboratory/aiai-skills                         # install all skills
npx skills add AIAI-Laboratory/aiai-skills --skill <skill-name>    # install one skill

Install manually

Clone the repository, then copy the skill you need into the agent's project-level skills directory:

git clone https://github.com/AIAI-Laboratory/aiai-skills.git
cd aiai-skills
mkdir -p /path/to/project/.agents/skills
cp -R skills/<skill-name> /path/to/project/.agents/skills/

The common destination is .agents/skills/; use these project-level paths when an agent has a native alternative:

Agent	Destination
Codex, Antigravity, and OpenCode	`.agents/skills/<skill-name>/`
Claude Code	`.claude/skills/<skill-name>/`
OpenCode native directory	`.opencode/skills/<skill-name>/`
Cursor	`.cursor/skills/<skill-name>/`

See docs/getting-started.md for user-level locations and alternatives.

Install from a Marketplace or Plugin

Use packaged installation when you want the repository to manage and update the complete skill collection:

Agent	Install
Claude Code	Add this repository as a marketplace, then install `aiai-skills@aiai-skills`
Codex	Add the marketplace, then run `codex plugin add aiai-skills@aiai-skills`
Antigravity	`agy plugin install https://github.com/AIAI-Laboratory/aiai-skills.git`

OpenCode can load the cloned repository through its root opencode.json; it can also use the manual .agents/skills/ or .opencode/skills/ installation above.

Skills Board

Skill	What It Does	Use When
codebase-replication	Learn advanced coding rules, syntax idioms, design patterns, file management, and tooling conventions from a source repository and apply them to the user's repository as a reusable rules file plus reusable knowledge pack.	Use this whenever the user says learn from this repo, copy style from, apply patterns from GitHub, modernize my code like X, replicate codebase, or gives a source repo path/URL plus a target repo. Always use for repo-to-repo style transfer even if they don't say skill.

How Skills Work

Skills use progressive disclosure so agents receive only the context needed for the current task:

┌──────────────────────────────────────────────────────────┐
│  SKILL.md                                                │
│                                                          │
│  ┌─ Frontmatter ──────────────────────────────────────┐  │
│  │ name: lowercase-hyphen-name                        │  │
│  │ description: What it does. Use when [user intent]. │  │
│  └────────────────────────────────────────────────────┘  │
│  Contributor Credit     → Contributed by @github-handle  │
│  Goal & Boundaries      → Outcome and scope, if needed   │
│  Workflow / Decisions  → Actions, defaults, and routing  │
│  Resource Routing      → Conditional supporting files    │
│  Constraints & Gotchas → Non-obvious rules and traps     │
│  Examples / Templates  → Clarify difficult cases         │
│  Verification          → Observable evidence             │
│  Output Contract       → Exact shape, when required      │
└──────────────────────────────────────────────────────────┘

A SKILL.md has required frontmatter, and each new skill credits its original contributor below the title. The body has no mandatory template: authors select only the sections that improve execution and arrange them in the order the agent needs them.

Design principles:

Precise discovery. Frontmatter says what the skill does and which user intents should activate it.
Useful instructions. The body captures reusable procedures, decision criteria, and non-obvious constraints rather than generic background.
Calibrated control. Flexible tasks get outcomes and defaults; fragile operations get exact ordering, checkpoints, and stopping conditions.
Progressive disclosure. Core guidance stays in SKILL.md; conditional detail lives in focused supporting files with explicit routing.
Observable verification. Completion is established by commands, artifacts, or externally visible behavior—not by wording that merely sounds correct.

Project Structure

.
├── skills/                         # Reusable skill packages
│   └── <skill-name>/
│       ├── SKILL.md
│       └── references/             # Optional supporting material
├── .claude-plugin/                 # Claude marketplace metadata
├── .codex-plugin/                  # Codex compatibility metadata
├── docs/                           # Authoring and installation guides
├── scripts/                        # Repository automation and validation
├── plugin.json                     # Portable Agent Plugins manifest
├── opencode.json                   # OpenCode skill-path configuration
└── README.md

Contributors

No contributors yet.

Contributing

Contribute skills that capture reusable, non-obvious expertise from real work. A good skill is focused enough to activate reliably, gives the agent only the guidance the task requires, calibrates precision to risk, and proves completion with observable evidence; extend an existing skill when it already owns the workflow.

See CONTRIBUTING.md and docs/skill-anatomy.md.

License

MIT — see LICENSE.