A curated collection of agent skills for coding agents designed to enhance the workflows of Software Engineers (and, let's be honest, me primarily).
Skills provide reusable, domain-specific knowledge and workflows that agents can use when relevant. Works with Cursor, Claude Code, Codex, OpenCode, Antigravity, GitHub Copilot, Amp, Roo Code, and other agent harnesses that support the Agent Skills standard.
Skills are modular, self-contained packages that extend agent capabilities with:
- Specialized workflows for specific domains
- Tool integrations and instructions
- Domain expertise and best practices
- Bundled resources (scripts, references, assets)
Skills use progressive disclosure - keeping SKILL.md files under 200 lines with detailed content in references/ directories loaded on-demand.
Install all skills globally (to ~/.cursor/skills/, ~/.claude/skills/, etc. depending on agent):
npx skills add siviter-xyz/dot-agent --global --agent cursorBy default, npx skills will prompt before installing or updating skills. You can add --yes (or -y) to skip confirmations.
Install individual skills by name:
# Install a single skill
npx skills add siviter-xyz/dot-agent --skill typescript --global
# Install multiple specific skills
npx skills add siviter-xyz/dot-agent --skill typescript --skill python --skill context-engineering --globalSee all available skills in this repository without installing:
npx skills add siviter-xyz/dot-agent --listTo discover skills across the broader ecosystem, use the built-in skills find command (powered by the internal find-skills helper skill; you don't need to install it separately):
npx skills find dot-agentTarget specific agent harnesses:
# Install to Cursor only
npx skills add siviter-xyz/dot-agent --agent cursor --global
# Install to multiple agents
npx skills add siviter-xyz/dot-agent --agent cursor --agent claude-code --globalInstall skills to your project (committed in repo):
# Install all skills to project
npx skills add siviter-xyz/dot-agent
# Install specific skill to project
npx skills add siviter-xyz/dot-agent --skill psiFor more options, see the skills CLI documentation.
When working on dot-agent locally, sync all skills to your local agent installation:
# Sync all skills from local dot-agent repo to Cursor
npx skills add "$(pwd)" --global --agent cursor
# Sync to multiple agents
npx skills add "$(pwd)" --global --agent cursor --agent claude-code
# Sync specific skills only
npx skills add "$(pwd)" --skill python --skill typescript --global --agent cursorThis is useful when developing or modifying skills locally before publishing.
See skills/README.md for a complete list of available skills.
For project-specific rules and commands, commit them in your repository:
- AGENTS.md: Project root or subdirectories for inline rules (works across agent harnesses, PREFERRRED)
- Rules:
.cursor/rules/,.claude/rules/,.codex/rules/, etc. in your repo - Commands:
.cursor/commands/,.claude/commands/,.codex/commands/, etc. in your repo
Keep project-specific content small and focused. Use skills for reusable knowledge.
All skills follow the progressive disclosure principle:
- SKILL.md < 200 lines - Core principles and when to use
- references/ - Detailed documentation loaded on-demand
- scripts/ - Executable code for deterministic tasks
- assets/ - Files used in output (templates, etc.)
This keeps context windows clean while providing access to specialized capabilities when needed.
Skills are designed to be:
- Focused - Each skill covers a specific domain
- Reusable - Work across projects and teams
- Maintainable - Clear structure with progressive disclosure
When contributing:
- Keep SKILL.md under 200 lines
- Use references/ for detailed content
- Follow existing skill structure and the Agent Skills specification
- Prefer agent-harness agnostic skills; if a skill is environment-specific, document this clearly in its frontmatter and/or README
- For scripts inside skills:
- Prefer uv scripts (PEP 723
# /// scriptmetadata with#!/usr/bin/env -S uv run --script) or plain Bash for simple glue - Aim for cross-platform behavior (Linux, macOS; avoid hard-coding shell- or OS-specific paths)
- Recommend installing
uvlocally and running scripts viauv runwhere appropriate
- Prefer uv scripts (PEP 723
- Test installation and usage