sayeed1999-agent-skills

Battle Tested Engineer

Instructions for your AI coding agent — not a library you import or an app you run.

This repo teaches your AI agent how to write and review code: ask before guessing, change only what was asked, prefer small, surgical fixes, and ship a working minimal version first.

Supported by Claude, Cursor, Codex, Gemini and other agent platforms.

This work was shaped by knowledge from earlier skill repositories such as andrej-karpathy-skills.

Quick Start Guide

1. Clone this repo:

  git clone https://github.com/sayeed1999/agent-skills.git
  cd agent-skills

2. Copy the skill into your agent skills dir:

User-wide:

mkdir -p ~/.agents/skills
cp -r skills/battle-tested-engineer ~/.agents/skills/

Project-scoped:

  mkdir -p .agents/skills
  cp -r skills/battle-tested-engineer .agents/skills/

3. In your AI agent, try a real prompt — e.g.,

/battle-tested-engineer explain the codebase

That's it!

What it does

Ten principles for writing, reviewing, and refactoring code — plus a clear, concise output style for agent interactions (code, commits, and PR descriptions remain normal prose).

• Think before code: state assumptions, avoid guessing.
• Simplicity: prefer minimal, maintainable changes.
• Architecture boundaries: keep business rules central.
• Surgical change: touch only what's requested.
• Data vs UI: separate data handling from rendering.
• Goal-driven execution: require verifiable goals and tests.
• Tests: small, high-value tests that protect behavior.
• Propose spec when needed: add brief PRDs/ADRs for big changes.
• Ship then improve: deliver minimal correct work, iterate.
• Gut-check before output: ensure diffs are proportionate and assumptions explicit.

Install

Tool	Method	Recommended?
Any agent	Copy skills/battle-tested-engineer/ to ~/.agents/skills/ or .agents/skills/	Yes (on-demand)
Cursor	Copy skills/battle-tested-engineer/ to ~/.cursor/skills/	Yes (on-demand)
Cursor	Copy .cursor/rules/battle-tested-engineer.mdc to project .cursor/rules/	Optional (always-on)
Claude Code	Plugin marketplace (see below)	Yes (global)
Claude Code	curl CLAUDE.md into project root	Per-project
Codex	Copy skill to ~/.codex/skills/ or .agents/skills/	Yes (on-demand)
Codex	curl AGENTS.md into project root	Per-project

Generic — .agents (any agent)

User-wide:

mkdir -p ~/.agents/skills
cp -r skills/battle-tested-engineer ~/.agents/skills/

Project-scoped:

mkdir -p .agents/skills
cp -r skills/battle-tested-engineer .agents/skills/

Claude Code — Plugin (recommended)

From within Claude Code, add the marketplace:

/plugin marketplace add sayeed1999/agent-skills

Then install:

/plugin install agent-skills@agent-skills

Claude Code — Per-project CLAUDE.md

New project:

curl -o CLAUDE.md https://raw.githubusercontent.com/sayeed1999/agent-skills/main/CLAUDE.md

Existing project (append):

echo "" >> CLAUDE.md
curl https://raw.githubusercontent.com/sayeed1999/agent-skills/main/CLAUDE.md >> CLAUDE.md

Cursor — On-demand skill (recommended)

cp -r skills/battle-tested-engineer ~/.cursor/skills/

See CURSOR.md for always-apply rule setup and tradeoffs.

Codex — On-demand skill (recommended)

Global:

cp -r skills/battle-tested-engineer ~/.codex/skills/

Repo-scoped:

mkdir -p .agents/skills
cp -r skills/battle-tested-engineer .agents/skills/

See CODEX.md for AGENTS.md per-project install.

How to use it

After install, you do not invoke the skill manually. Just open a project and talk to your agent as usual.

1. Install using one of the methods in Install above (pick your tool; on-demand skill is recommended).
2. Open a codebase in Cursor, Claude Code, or Codex.
3. Ask for work the way you already would — e.g. "fix the login 500", "review this diff", "add tests for checkout", "refactor this component".
4. Let the agent apply the guidelines — it should clarify ambiguous asks, keep diffs small, and write minimal code. See EXAMPLE.md for good vs bad behavior.

On-demand skill: loaded when the task fits (refactor, review, tests, UI). General questions may not trigger it.

Always-on rule or CLAUDE.md / AGENTS.md: guidelines apply every message in that project, including the concise output style during code work.

Optional: merge project-specific rules into the same files (see Customization).

How to know it's working

• Fewer unnecessary changes in diffs; changes are narrowly scoped.
• Clarifying questions appear before implementation.
• Characterization tests for risky behavior changes.
• Concise, human explanations and clear specs for decisions.
• Clean, minimal PRs with focused diffs.

Customization

These guidelines merge with project-specific instructions. Add sections like:

## Project-Specific Guidelines

- Use TypeScript strict mode
- All API endpoints must have tests
- Follow error handling in `src/utils/errors.ts`

Tradeoff note

These guidelines bias toward caution over speed. The skill applies a concise, no-friction output style for code work; it is opt-in when using on-demand installs or always-apply rules in tools like Cursor or per-project CLAUDE.md / AGENTS.md.

Use judgment for trivial tasks — not every edit requires full process.

Repository structure

agent-skills/
├── .claude-plugin/          # Claude Code plugin manifests
├── .cursor/rules/           # Optional always-apply Cursor rule
├── skills/                  # Canonical skill definitions
│   └── battle-tested-engineer/
│       └── SKILL.md
├── CLAUDE.md                # Claude Code per-project install
├── AGENTS.md                # Codex per-project install
├── CURSOR.md                # Cursor setup guide
├── CODEX.md                 # Codex setup guide
└── EXAMPLE.md               # Good vs bad agent behavior

License

MIT