sanjay3290/ai-skills: 24 Portable Skills for Claude, Cursor & Codex

sanjay3290/ai-skills: 24 Portable Skills for Claude, Cursor & Codex

AI coding assistants have become standard equipment for most developers, but there's a growing friction: every agent speaks a slightly different dialect. Claude Code, Cursor, Gemini CLI, OpenAI Codex CLI—each has its own plugin format, its own configuration path, its own idea of how to extend capabilities. If you switch tools or work across teams using different agents, you end up rebuilding the same integrations from scratch. That's the concrete problem sanjay3290/ai-skills addresses. This open-source collection provides 24 portable, cross-platform skills that work across all major AI clients supporting the Agent Skills Standard. Instead of locking your tooling into one ecosystem, you install once and use everywhere.

The repository has gained 343 GitHub stars and 37 forks since its release, with the most recent commit on July 9, 2026. It's written primarily in Python^{↗ Bright Coding Blog} and released under the Apache License 2.0. This article walks through what the project offers, how to install and configure it, and where it fits in a modern development workflow.

What is sanjay3290/ai-skills?

sanjay3290/ai-skills is a curated collection of 24 cross-platform agent skills for AI coding assistants. Each skill is a self-contained module that extends what your AI agent can do—query databases, generate images, manage cloud resources, send messages, or delegate tasks to other AI systems. The project is maintained by sanjay3290 and built around the open Agent Skills Standard, which defines a common format for skill packaging and discovery.

The repository sits at the intersection of several active trends in developer tooling: the rise of agentic AI workflows, the fragmentation of AI client ecosystems, and the demand for vendor-neutral infrastructure. Rather than betting on Claude Code or Cursor or Gemini CLI winning the market, this project treats them as interchangeable targets. Skills install into ~/.claude/skills/, ~/.cursor/skills/, ~/.gemini/skills/, or project-local equivalents depending on your agent.

The 24 skills span six broad domains: databases (PostgreSQL^{↗ Bright Coding Blog}, MySQL^{↗ Bright Coding Blog}, MSSQL), messaging and collaboration (Telegram, Google Chat, Gmail), research and content generation (Deep Research, Imagen, NotebookLM), task delegation (Jules, Manus, Grok Build), DevOps^{↗ Bright Coding Blog} and project management (Azure DevOps, Atlassian, Apple Container), and Google Workspace (Docs, Sheets, Slides, Drive, Calendar). This breadth reflects a pragmatic philosophy: developers already have the tools they need, and AI agents should connect to them rather than replace them.

Key Features

Cross-platform agent compatibility. The project explicitly supports six major AI clients: Claude Code, Gemini CLI, Google Antigravity, Cursor, OpenAI Codex CLI, and Goose. Each has documented installation paths and directory mappings. The npx skills CLI auto-detects your agent or lets you target specific ones with -a flags.

Standardized skill structure. Every skill follows the same layout: SKILL.md with YAML frontmatter for agent instructions, README.md for human documentation, optional requirements.txt, .env.example, and a scripts/ directory with executable code. This predictability makes skills easy to inspect, modify, and contribute back.

Multiple installation methods. You can install via Claude Code's native plugin marketplace (/plugin marketplace add sanjay3290/ai-skills), the npx skills package manager, or manual git clone/symlink. Global (-g) and project-local installs are both supported.

Defense-in-depth security for database access. The database skills (PostgreSQL, MySQL, MSSQL) are read-only by design with multiple protection layers: session-level write protection, query validation, and explicit connection configuration via connections.json rather than environment variables.

Secure credential storage. Google Workspace skills use system keyring integration (macOS Keychain, Windows Credential Locker, Linux Secret Service API) instead of plaintext files. OAuth flows are handled through a shared cloud function derived from the official Google Workspace MCP server.

Async task delegation. Skills for Jules, Manus, and Grok Build enable your primary agent to dispatch work to specialized AI systems and retrieve results later—useful for long-running tasks like comprehensive research, bug fixes, or multi-file refactors.

TTS and podcast generation. Both ElevenLabs and Google Cloud TTS skills support narrating documents (PDF, DOCX, MD, TXT) and generating two-host podcast formats, with multi-chunk processing for long content.

Use Cases

Multi-database exploration without context switching. A backend engineer working across PostgreSQL production, MySQL legacy, and MSSQL analytics systems can query all three through their AI agent without leaving the terminal. The read-only design provides safety guardrails for production access.

Research-driven technical decisions. The Deep Research skill uses Gemini's research agent to compile multi-step investigations—comparing frameworks, analyzing competitive landscapes, or tracking technology developments. Results arrive in 2-10 minutes, suitable for architecture decision records or spike documentation.

Async bug fixing and test generation. The Jules skill delegates coding tasks to Google's Jules AI agent. A developer can dispatch "fix authentication bug in src/auth.js" or "add unit tests for API module," continue their own work, and pull results when Jules completes asynchronously.

Team coordination through existing channels. The Telegram skill sends alerts and files via bot, with ask-and-wait approval flows using inline buttons. The Google Workspace suite (Chat, Docs, Sheets, Calendar) integrates with corporate infrastructure without requiring the full Google Workspace MCP server deployment.

Headless CI/CD management. The Azure DevOps skill exposes 99 tools across 13 domains—work items, repos, PRs, pipelines, test plans, security alerts, deployment approvals. DevOps engineers can query status, trigger builds, or check pending approvals conversationally.

Installation & Setup

Claude Code Plugin Marketplace (Fastest)

/plugin marketplace add sanjay3290/ai-skills
/plugin install ai-skills@ai-skills

This installs all 24 skills as a single plugin. Updates use /plugin marketplace update ai-skills.

npx skills CLI (Any Agent)

The npx skills package manager works with 40+ agents. Install without global dependencies:

# Browse available skills
npx skills add sanjay3290/ai-skills --list

# Install single skill (auto-detects agent)
npx skills add sanjay3290/ai-skills --skill postgres

# Install multiple skills
npx skills add sanjay3290/ai-skills --skill postgres --skill mysql --skill mssql

# Install all 24 skills
npx skills add sanjay3290/ai-skills --all

Target specific agents explicitly:

# Claude Code specifically
npx skills add sanjay3290/ai-skills --skill postgres -a claude-code

# Multiple agents simultaneously
npx skills add sanjay3290/ai-skills --skill postgres -a claude-code -a gemini-cli -a cursor

# All agents
npx skills add sanjay3290/ai-skills --all -a '*'

Global vs Project Scope

# Global — available in all projects
npx skills add sanjay3290/ai-skills --skill imagen -g

# Project-local (default) — scoped to current repo
npx skills add sanjay3290/ai-skills --skill imagen

Skill Management

# List installed skills
npx skills list

# Check for updates
npx skills check

# Update all
npx skills update

# Remove specific skill
npx skills remove postgres

# Remove all from specific agent
npx skills remove --skill '*' -a cursor

Manual Installation (Alternative)

# Clone entire repository
git clone https://github.com/sanjay3290/ai-skills.git ~/.claude/skills/ai-skills

# Copy individual skill
cp -r skills/postgres ~/.claude/skills/

# Symlink for development
ln -s /path/to/ai-skills/skills/postgres ~/.claude/skills/postgres

Replace ~/.claude/skills/ with your agent's directory as needed.

Real Code Examples

Database Connection Configuration

The PostgreSQL, MySQL, and MSSQL skills require a connections.json file in the skill directory:

{
  "production": {
    "host": "db.example.com",
    "port": 5432,
    "database": "app_production",
    "user": "readonly_user",
    "password": "${PROD_DB_PASSWORD}"
  },
  "staging": {
    "host": "staging.internal",
    "port": 5432,
    "database": "app_staging",
    "user": "dev_user",
    "password": "${STAGING_DB_PASSWORD}"
  }
}

Install the appropriate driver for your database:

pip install psycopg2-binary     # PostgreSQL
pip install mysql-connector-python  # MySQL
pip install pymssql              # MSSQL

The skill reads this configuration and enforces read-only access at the session level. The ${VAR} syntax allows environment variable substitution without storing secrets in version control.

Google Workspace Authentication

Each Google Workspace skill requires one-time OAuth authentication with secure token storage:

# Install keyring dependency (one-time)
pip install keyring

# Authenticate per skill (opens browser)
python ~/.claude/skills/ai-skills/skills/google-chat/scripts/auth.py login
python ~/.claude/skills/ai-skills/skills/google-docs/scripts/auth.py login
python ~/.claude/skills/ai-skills/skills/gmail/scripts/auth.py login

Tokens are stored in the system keyring—macOS Keychain, Windows Credential Locker, or Linux Secret Service API. This avoids the common anti-pattern of .env files with long-lived OAuth tokens. The authentication flow uses the same OAuth infrastructure as the official Google Workspace MCP server.

Skill Structure Template

Every skill follows this standardized layout from the Agent Skills Standard:

skill-name/
├── SKILL.md              # Required: Instructions for the AI agent
├── README.md             # Human documentation
├── requirements.txt      # Dependencies (if any)
├── .env.example          # Environment variable template
└── scripts/              # Executable scripts
    └── main.py

The SKILL.md file uses YAML frontmatter to declare metadata:

---
name: postgres
description: "Read-only PostgreSQL queries with defense-in-depth security"
---

# Instructions for the AI agent

When the user asks about database data, use the configured connections.json
to connect. Validate all queries are SELECT statements before execution.
Return results in markdown↗ Smart Converter tables for readability.

This structure makes skills self-documenting and portable across agents. The agent reads SKILL.md to understand when and how to invoke the skill; humans read README.md for setup and configuration.

Advanced Usage & Best Practices

Prefer project-local installs for team consistency. While global installs are convenient for personal workflows, committing .claude/skills/ or .cursor/skills/ to your repository ensures teammates get the same capabilities without manual setup. The npx skills CLI respects project-local skills first.

Segment database credentials by environment. The connections.json format supports multiple named connections. Name them explicitly (production, staging, analytics) to prevent accidental cross-environment queries. The read-only enforcement is a safety net, not a substitute for principle-of-least-privilege database users.

Budget for Deep Research costs. The README notes Deep Research tasks cost $2-5 per query and take 2-10 minutes. This is not a free replacement for web search—reserve it for decisions where comprehensive synthesis justifies the expense and latency.

Use async delegation for parallel work streams. Jules, Manus, and Grok Build skills all support fire-and-forget patterns. A productive workflow: dispatch a Jules task for test generation, continue coding elsewhere, then ask "Check my Jules sessions" when you need results. This mirrors how senior engineers delegate to junior team members.

Monitor Google Workspace token expiration. The OAuth tokens stored in keyring are refreshable but not permanent. If a skill suddenly fails with authentication errors, re-run the auth.py login script for that specific skill rather than debugging code.

Comparison with Alternatives

sanjay3290/ai-skills occupies a middle ground: more integrated than ad-hoc MCP servers, more portable than agent-native tools. The trade-off is setup complexity—you configure each skill individually rather than getting batteries-included defaults. For teams committed to a single agent, native tools may be smoother. For polyglot environments or organizations avoiding vendor lock-in, the portability justifies the configuration overhead.

The project also complements rather than replaces the Google Workspace MCP server—the README explicitly positions its Google skills as "lightweight alternatives" for cases where full MCP deployment is excessive.

FAQ

Q: Does this work with GitHub Copilot? A: The npx skills CLI lists github-copilot as a supported agent with .github/skills/ paths, though the README's detailed documentation focuses on Claude Code, Gemini CLI, Cursor, Codex, and Goose.

Q: Are the database skills safe for production? A: They enforce read-only access with query validation and session-level write protection, but you should still use dedicated read-only database users and review connections.json access controls.

Q: What's the license? A: Apache License 2.0, permitting commercial use, modification, and distribution with attribution.

Q: How do I update skills when new versions release? A: Claude Code users run /plugin marketplace update ai-skills. CLI users run npx skills update.

Q: Can I contribute new skills? A: Yes—fork the repository, create a skill in skills/your-skill-name/ with SKILL.md and README.md, and submit a pull request.

Q: Do I need Python installed? A: Yes, for skills with Python scripts. The requirements.txt in each skill specifies dependencies. Some skills additionally need Node.js (Jules) or system tools like ffmpeg (TTS skills).

Q: Why does Google Workspace require a business account? A: Google's API restrictions—these skills use Workspace APIs that Google does not enable for personal Gmail accounts.

Conclusion

sanjay3290/ai-skills solves a real and growing problem: AI agent fragmentation. With 343 stars and active maintenance through mid-2026, it has found traction among developers who work across multiple AI clients or want to avoid ecosystem lock-in. The 24 skills cover enough ground—databases, research, task delegation, messaging, DevOps, and office productivity—that most teams will find several immediately useful.

This project is best suited for: developers using multiple AI agents (Claude Code for deep work, Cursor for IDE integration, Gemini CLI for Google ecosystem tasks); teams wanting standardized tooling across personal and project contexts; and organizations prioritizing vendor independence over native integration polish.

It's less ideal for: developers committed to a single agent with robust native tools; users wanting zero-configuration setup; or those unwilling to manage per-skill authentication flows.

The open standard approach, Apache 2.0 licensing, and active contribution guidelines suggest this will continue evolving with the broader agent ecosystem. If you're investing time in AI-assisted development, the portability sanjay3290/ai-skills provides is worth the upfront configuration.

Explore the full collection at https://github.com/sanjay3290/ai-skills.