Version your team's AI coding standards as code. Keep your Cursor skills and rules in git, auto-sync them into every developer's editor, and review changes to them like any other code.
agent-standards-as-code is a small, opinionated starter template for teams who use AI coding agents across many repos and want everyone's agent to follow the same standards — without copy-pasting instruction files around and watching them drift.
This is v0.1 — the core mechanism, targeting Cursor. Multi-tool support (Claude Code, Copilot), cross-platform scripts, a freshness loop, and system-level memory arrive in later versions (see Roadmap).
A single instructions file is fine for one repo and one person. With a team and a fleet of repos it breaks down:
- It doesn't scale — every repo gets a slightly different copy.
- It drifts — nobody knows which version is current.
- It's unversioned — no ownership, no PR review of your "AI brain."
agent-standards-kit makes your standards a single, git-versioned source of truth that syncs into every developer's Cursor automatically.
# 1. Click "Use this template" above, or clone:
git clone https://github.com/prathakmalik/agent-standards-kit.git
cd agent-standards-kit
# 2. Run setup (no admin needed):
./scripts/setup.ps1
# Blocked by execution policy? -> powershell -ExecutionPolicy Bypass -File .\scripts\setup.ps1
# 3. Restart Cursor. Your skills + rules are now active in every workspace.setup.ps1 does three things:
- Syncs
skills/into~/.cursor/skills/via directory junctions (no admin). - Syncs
rules/into~/.cursor/rules/via content-compared file copies. - Installs git hooks so a future
git pull/checkout/rebasere-syncs automatically.
agent-standards-kit/
├── skills/ # task workflows (SKILL.md format)
│ ├── generate-readme/
│ └── review-changes/
├── rules/ # standing guidance (Cursor .mdc)
│ ├── coding-standards.mdc
│ └── pr-checklist.mdc
├── scripts/ # setup / sync / hooks (PowerShell)
├── examples/ # a de-identified worked example
└── .github/ # issue + PR templates, CI lint
- Skills = "here's how to do task Y" (dynamically loaded).
- Rules = "always do X" (standing behavior).
.mdcfiles supportalwaysApplyandglobs.
flowchart LR
SoT["Git repo (source of truth)<br/>skills/ + rules/"] --> Sync["setup.ps1<br/>sync + git hooks"]
Sync -->|junction| Skills["~/.cursor/skills/"]
Sync -->|copy| Rules["~/.cursor/rules/"]
Skills --> Cursor["Cursor (every workspace)"]
Rules --> Cursor
Pull["git pull / checkout / rebase"] --> Hooks[".git/hooks"] --> Sync
Because rules sync to the user-level ~/.cursor/rules/, they apply across every workspace automatically. Update the repo, git pull, and everyone's standards move together.
| Version | Theme |
|---|---|
| v0.1 (this one) | Core mechanism, Cursor, sample skills/rules |
| v0.2 | Multi-tool (Claude Code, Copilot) + cross-platform (bash + PowerShell) |
| v0.3 | The nudge -> skill -> PR freshness loop + "make it yours" |
| v0.4 | System-level memory: ecosystem map + post-mortem records |
PRs welcome — see CONTRIBUTING.md. CI validates SKILL.md frontmatter.
MIT.