Skip to content

sarathsomana/tech-skills

Tech Skills

badge badge badge badge

Structured, agent-agnostic skills that turn any AI coding assistant into an expert engineering team.

A skill is a reusable instruction set that guides an AI agent through complex tasks using defined phases, specialized personas, and quality gates. Each skill is a single Markdown file with YAML frontmatter — no vendor lock-in, no runtime dependencies. Install a skill, and your agent gains a structured workflow it can follow to produce reliable, auditable results.

Why Tech Skills?

Without Skills With Tech Skills
AI generates code in a single pass with no structure Agent follows defined engineering phases with quality gates between each step
Debugging is ad-hoc — "try this, try that" Binary-search isolation with hypothesis testing and artifact-gated validation
Architecture decisions are based on pattern matching First-principles deconstruction that challenges assumptions to their root constraints
UI feedback is subjective and inconsistent Aesthetic audits scored against a formal design system rubric
Security reviews depend on what the model remembers Red-team threat modeling with structured vulnerability classification

Quick Start

Paste this into your AI coding agent (Claude Code, Gemini CLI, Cursor, Windsurf, Cline, or any agent that reads Markdown):

Install skills from https://raw.githubusercontent.com/sarathsomana/tech-skills/main/INSTALL.md

The agent will read the instructions, detect your environment, and install the skills automatically. See the full installation guide for other options.

Once installed, try:

"Use first-principles-design to architect my new API."

Available Skills

Foundation and Meta-Architecture

Skill Description Use When
Skill Architect Designs, scaffolds, and validates new skills using the Phased State Machine architecture You need to create a new skill from scratch
Base State-Machine Runtime Standardized templates and runtime rules for phased state-machine skills You are building on the framework and need the base contract
First Principles Design Challenges assumptions through structured deconstruction until fundamental constraints are identified, then reconstructs the design from those truths You are starting a new architecture or questioning an existing one

Core Engineering

Skill Description Use When
Implementation Engineer Ingests architectural blueprints and executes phased coding, testing, and integration You have a design and need to turn it into working code
Root-Cause Isolator Binary-search style debugging with hypothesis testing and automated patch generation You have a failing system and need to find the exact cause
Legacy Modernizer Extracts logic from legacy code and refactors incrementally with functional equivalence guarantees You need to modernize old code without breaking existing behavior
Graphify v2 Generates knowledge graphs from codebases for semantic understanding, dependency analysis, and architectural auditing You need to understand a large or unfamiliar codebase

Visual and Experience

Skill Description Use When
UX Guardian Aesthetic audits, CSS refactoring, and animation polishing against design system standards You want to audit or improve the visual quality of a UI
Component Wizard High-fidelity UI component generation from screenshots or design specifications You need to build a component that matches a reference design

Hardening and Security

Skill Description Use When
Security Sentinel Red-team auditing, vulnerability patching, and threat modeling You need to verify a system is secure before production
Perf-Optimus Frontend and backend performance optimization with empirical before/after validation You need to identify and fix performance bottlenecks
Test Engineer Test strategy design, test authoring, execution, and failure diagnosis You need a comprehensive test plan or want to improve test coverage

Ecosystem and Integration

Skill Description Use When
Multi-Agent Orchestrator Task decomposition and handover logic between different agents using shared skills You need multiple agents to collaborate on a single task
Skill Registry Searchable index of community-contributed skills for discovery, validation, and installation You want to find or publish skills

How It Works

Each skill follows the Phased State Machine architecture: a defined sequence of phases, each governed by a specialized persona, with transitions gated by concrete artifacts rather than open-ended reasoning.

graph LR
    A[Your AI Agent] -->|reads| B["SKILL.md"]
    B --> C{State Machine}
    C -->|Phase 1| D["Scout — gather context"]
    C -->|Phase 2| E["Visionary — design"]
    C -->|Phase 3| F["Pragmatist — implement"]
    C -->|Phase 4| G["Critic — validate"]
    D -->|artifact| E
    E -->|artifact| F
    F -->|artifact| G
    G -->|pass| H["Complete"]
    G -->|fail| F
Loading

Key principles:

  • Artifact-gated transitions. A phase produces a tangible output (Markdown, JSON, code, or test results). The next phase uses that artifact as its source of truth. No hand-waving.
  • Composite personas. Each phase activates a different cognitive temperament — a Scout gathers context, a Pragmatist builds, a Critic audits. This prevents a single mode of thinking from dominating the workflow.
  • Deterministic execution. State transitions are gated on physical assertions (exit codes, file existence, test results), not on the model's self-assessment.
  • Agent-agnostic format. Skills are plain Markdown with YAML frontmatter. They work with any agent that can read a file.

For the full design rationale, see Philosophy: Phased State Machines.

Installation

Option A: Install via Your AI Agent (Recommended)

Paste this into your coding agent:

Install skills from https://raw.githubusercontent.com/sarathsomana/tech-skills/main/INSTALL.md

The agent reads INSTALL.md, detects your environment, and creates symlinks to the correct discovery path. Supports both global and project-level installation.

Option B: Install via Shell Script

# Default: installs to ~/.agents/skills/
curl -sL https://raw.githubusercontent.com/sarathsomana/tech-skills/main/install.sh | bash

# Agent-specific: installs to the correct path for your agent
curl -sL .../install.sh | bash -s -- --agent claude
curl -sL .../install.sh | bash -s -- --agent gemini --scope project

Supported agents: claude, gemini, antigravity, cursor, windsurf, cline, copilot, generic. Run install.sh --help for all options.

Option C: Manual Installation

git clone https://github.com/sarathsomana/tech-skills.git ~/.tech-skills
ln -s ~/.tech-skills/skills/* ~/.agents/skills/

Local Development

If you have cloned the repository and want to install skills locally for testing:

./install-local.sh

This creates symlinks from skills/ to .agents/skills/ within the repository.

Usage Examples

Category Example Prompt
Architecture "Use first-principles-design to evaluate whether we actually need a microservices architecture."
Implementation "Use implementation-engineer to build the API layer from this design document."
Debugging "Use root-cause-isolator to find why this integration test is failing intermittently."
Modernization "Use legacy-modernizer to refactor this jQuery module to modern JavaScript."
UX "Use ux-guardian to audit the visual quality of my dashboard page."
Security "Use security-sentinel to red-team my authentication and session management."
Performance "Use perf-optimus to profile and optimize this page's load time."
Testing "Use test-engineer to design a test strategy for the payments module."

Contributing

We welcome contributions — new skills, improvements to existing ones, and bug reports.

  • Add a new skill: Fork the repo, create a directory under skills/, add a SKILL.md with YAML frontmatter, and submit a PR. See CONTRIBUTING.md for details.
  • Improve a skill: Submit a PR updating the relevant SKILL.md or associated files.
  • Report an issue: Use the bug report or skill request templates.
  • Code of Conduct: CODE_OF_CONDUCT.md

References

Attributions

License

MIT

About

Open-Source Skills that are Structured and agent-agnostic. Turn any AI coding assistant into an expert engineering team.

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages