Pragmatic Agent Skills

An open collection of production-minded agent skills.

中文说明

This repository focuses on one thing: reusable skills that solve real tasks with explicit rules, fallbacks, and operational boundaries. It is not a prompt dump, and it is not a generic AI toolbox.

Why This Repo Exists

Most public AI repos have one of two problems:

• they are broad but shallow
• they look useful, but do not carry enough execution detail to be reused safely

Pragmatic Agent Skills is opinionated in the opposite direction:

• each skill is narrow and concrete
• each skill defines trigger conditions and non-goals
• important fallback behavior is written down
• scripts and templates are included when the skill needs them

If you build with Cursor, Claude Code, Codex, Windsurf, or similar agent systems, the goal is simple: install a skill and get repeatable behavior instead of rewriting the same working pattern every week.

What Is In Here

Current skills:

Skill	What it does	Why it is interesting
JobJD-Greeting-Generator	Turns a JD and candidate facts into one copy-ready recruiting message	Strong fallback design around jd_url extraction instead of giving up early
leetcode-coach	Keeps public coaching logic separate from private LeetCode records	Clean split between shared skill logic and user-owned data

Browse the full index in skills/README.md.

Related Tools

self-improvement-skill started here, but it is now maintained as a standalone npm CLI because it installs hooks and manages runtime integration rather than behaving like a plain skill directory.

Install it with:

npx self-improvement-skill install

Repository: CAN1177/self-improvement-skill

Design Principles

Every skill in this repo should follow the same bar:

1. Solve one concrete problem.
2. State when the skill should and should not trigger.
3. Prefer deterministic workflow over vague guidance.
4. Preserve user-owned data boundaries.
5. Fail explicitly when confidence is low.
6. Stay useful as a public artifact, not just a personal note.

More detail: docs/design-principles.md

Quick Start

1. Pick a skill from skills/README.md
2. Read its README.md for the public contract
3. Read its SKILL.md for agent-facing rules
4. Reuse the included scripts/templates if the skill provides them

Setup notes: docs/getting-started.md

Generic Prompts vs Pragmatic Skills

Generic prompts	Pragmatic skills
broad instructions	narrow task scope
weak trigger boundaries	explicit trigger and non-goal definitions
one-shot output bias	repeatable workflow bias
hidden fallback behavior	documented fallback path
hard to reuse consistently	easier to install and reuse across agent runtimes

Repository Layout

.
├── skills/   # public, reusable agent skills
├── docs/     # repo-level principles and onboarding docs
└── tests/    # focused validation for selected skills

Why People Might Star This

• the skills are usable, not aspirational
• the repository shows how to package agent behavior cleanly
• the examples are grounded in real workflows
• the implementation style is narrow enough to copy

If this repo becomes valuable, it will be because each skill is a small, opinionated piece of working infrastructure.

Contributing

Issues and PRs are welcome, especially when they improve one of these:

• skill clarity
• fallback behavior
• examples
• validation
• installation ergonomics

Contribution notes: CONTRIBUTING.md

Changelog

Recent repository history: CHANGELOG.md