Skills for Agents

Composable, domain-specific instruction sets for AI agents

Documentation • Getting Started • SIP Framework • Contributing

Overview

Skills for Agents is a production-ready ecosystem of composable instruction sets that enable AI agents to handle multiple concerns simultaneously without conflict. Each skill owns one domain (voice, density, craft, process, or content) and works alongside any other skill through the Skills Interoperability Protocol (SIP).

Why Skills?

Traditional AI prompts are monolithic — one giant instruction set that tries to do everything. When you need multiple capabilities, they conflict. Skills solve this through:

• Domain separation — each skill owns one aspect (voice, density, craft, process, content)
• Explicit composition — skills declare what they yield to
• Conflict resolution — SIP Framework provides precedence rules

Result: skills that compose without breaking each other.

Quick Start

Installation

Via npx (recommended):

# Install all skills to all detected tools
npx skills-for-agents install

# Install to a specific tool
npx skills-for-agents install --tool claude
npx skills-for-agents install --tool cursor
npx skills-for-agents install --tool windsurf
npx skills-for-agents install --tool codex
npx skills-for-agents install --tool antigravity
npx skills-for-agents install --tool kiro
npx skills-for-agents install --tool zed --project
npx skills-for-agents install --tool cline --project
npx skills-for-agents install --tool aider --project
npx skills-for-agents install --tool copilot --project
npx skills-for-agents install --tool continue
npx skills-for-agents install --tool hermes

# Install to current project directory
npx skills-for-agents install --project

# Install specific skills only
npx skills-for-agents install --only caveman,blogger,slidify

# List available skills
npx skills-for-agents list

Via git clone (manual):

git clone https://github.com/IsNoobgrammer/skills-for-agents.git

Each skill is a self-contained folder with a SKILL.md file. Point your agent to load these files as system prompts.

Basic Usage

Invoke a skill:

/caveman full

Compose skills (layered):

/blog technical + /caveman lite
→ Blogger writes technical post, caveman compresses it

Compose skills (pipeline):

/postmortem → /compress
→ Postmortem generates report, compress shrinks it

Natural language:

"Write a blog about the UI incident, make it terse"
→ Auto-detects: blogger (voice) + caveman (density) + postmortem (content)

Available Skills

Voice

| Skill | Description | Triggers | |-------|-------------|----------| | Blogger | Authentic personal-voice writing. Raw, unpolished, stream-of-consciousness prose. | /blog, "write a blog", "write like me" |

Density

| Skill | Description | Triggers | |-------|-------------|----------| | Caveman | Ultra-terse communication. Cuts token usage ~75%. | /caveman, "talk like caveman", "less tokens" | | Compress | File compression. 4 intensity levels. Preserves all meaning. | /compress, "compress this", "reduce tokens" |

Craft

| Skill | Description | Triggers | |-------|-------------|----------| | Painter | Max pro UI/UX design. Animation, color, typography, accessibility. 20+ commands. | /painter, "make it look pro", "fix the ui" | | Harden | Production-harden code for 1M+ users. Caching, rate limiting, graceful shutdown. | "harden my code", "prepare for launch", "make it scalable" |

Process

| Skill | Description | Triggers | |-------|-------------|----------| | Memory | Persistent context engine. Daily journal rotation, manifest indexing, identity tracking. | Startup (mandatory), "I like X", "Here is my key" | | ML Engine | TPU-first ML research. Distributed training, MoE, Pallas kernels. 12+ commands. | /ml, TPU mentions, distributed training | | Planner | Project plans with PRDs, design docs, architecture flows, task breakdowns. | /plan, "create a PRD", "design the architecture" | | Postmortem | Blameless incident documentation. 5 Whys, action items. | /postmortem, "incident review", "what broke and why" | | Refactor | Restructure messy codebases into clean, modular architecture. | "refactor my project", "clean up my code", "split this file" | | Skill Creator | Meta-skill for creating, auditing, and improving other skills. | /create-skill, "make a skill", "turn this into a skill" | | Slidify | End-to-end PowerPoint generator. JSON specs, templates, auto speaker notes. | /slidify, "make a presentation", "create slides", "generate pptx" |

Content

| Skill | Description | Triggers | |-------|-------------|----------| | Documenter | Comprehensive documentation. Examples, guides, API references. | "document this", "write docs", "create documentation" | | Learn | Structured study plans, topic guides, exam prediction, and active recall. | /learn, "teach me X", "how do I learn Y" | | Researcher | Deep web research. Diverse sources, cross-referencing, synthesis. | "research X", "find info about Y", "what's the latest on Z" |

GPU Engineering

| Skill | Description | Triggers | |-------|-------------|----------| | Tritonify | Agent-driven GPU kernel optimization. Triton + CUDA. Profiling-guided iteration, 52+ papers, MoE/loss/attention kernels. | "optimize kernel", "write triton", "profile CUDA", "fuse operators", "speed up LLM ops" |

Analysis

| Skill | Description | Triggers | |-------|-------------|----------| | OSINT | Open Source Intelligence engine. Email, phone, username, domain recon. Breach checks, social media, infrastructure. | "osint", "recon", "background check", "investigate" | | Godmode | Jailbreak LLMs: Parseltongue, GODMODE, ULTRAPLINIAN. Compliance escalation engine. Ported from Hermes Agent. | "jailbreak", "godmode", "red team", "bypass guardrails", "uncensor" |

Key Features

🎯 Domain-Specific

Each skill owns one aspect — voice, density, craft, or process. No overlap, no conflict.

🔗 Composable

Skills work together seamlessly. Caveman + Blogger = terse posts in authentic voice.

📐 SIP Framework

Skills Interoperability Protocol ensures every skill knows how to compose with others.

⚡ Production-Ready

Battle-tested skills for ML research, documentation, UI/UX, incident response, and more.

🎨 18 Skills Included

Blogger, Caveman, Compress, Documenter, Godmode, Harden, Learn, Memory, ML Engine, OSINT, Painter, Planner, Postmortem, Refactor, Researcher, Skill Creator, Slidify, Tritonify.

🚀 Framework-Agnostic

Drop into any agent framework. Works with any LLM that supports system prompts.

Composition Examples

Terse Technical Writing

/blog technical + /caveman lite

Blogger writes technical post, caveman compresses it.

Production-Ready Code

/refactor → /harden

Refactor establishes structure, harden adds production patterns.

Comprehensive Docs

/documenter + /researcher

Researcher gathers context, documenter structures it into docs.

Incident Response

/postmortem → /compress

Postmortem generates report, compress shrinks it for storage.

ML Research

/ml-engine + /researcher

Researcher finds prior work, ml-engine implements experiments.

SIP Framework

The Skills Interoperability Protocol is the composability contract that makes multi-skill execution possible.

Core Concepts

Domains:

• Voice — tone, personality, vocabulary
• Density — token count, verbosity
• Craft — visual design, code quality
• Process — workflow steps, templates
• Content — substance being produced

Composition Modes:

• Layered — simultaneous application (/skill1 + /skill2)
• Pipeline — sequential processing (/skill1 → /skill2)
• Handoff — skill delegates to another
• Advisory — skill references another's principles

Precedence Rules:

1. Safety/Accuracy (always wins)
2. User's explicit instruction
3. Domain owner (authoritative in its domain)
4. Most recently invoked
5. Specificity (narrow scope beats broad)

Read full SIP specification →

Documentation

• Live Documentation — Complete guides, API references, examples
• Getting Started — Installation and basic usage
• SIP Framework — Composability contract and domain rules
• Creating Skills — Build your own skills
• Contributing Guide — How to contribute to the ecosystem
• Automation Bots — Automated PR review and validation

Contributing

We welcome contributions! Whether you're:

• Creating new skills
• Improving existing skills
• Fixing bugs
• Enhancing documentation
• Building tooling

Please read our Contributing Guide to get started.

Creating a New Skill

Use the skill-creator skill to scaffold a new skill with SIP compliance built in:

/create-skill

Or manually:

1. Create a new folder with your skill name (kebab-case)
2. Add a SKILL.md file with frontmatter declaring your domain
3. Follow the SIP Framework composability rules
4. Submit a PR with your skill

License

MIT License - see LICENSE for details.

Acknowledgments

Built on the principles of:

• Domain-Driven Design — clear boundaries, explicit contracts
• Unix Philosophy — do one thing well, compose freely
• Progressive Enhancement — works standalone, better together

Documentation • SIP Framework • Contributing

Made with ❤️ by Shaurya Sharthak