✂ snip

Ruthless efficiency mode for AI coding agents. Delete first. Build last.

Snip enforces a 7-rung deletion ladder on AI agents — starting with "can this be deleted entirely?" — and protects production paths from over-simplification with snip:prod annotations.

The problem snip solves

AI agents have a strong bias toward completeness. Ask for email validation, get a 27-line EmailValidator class with a regex that still rejects valid addresses. Ask for a cache, get a 120-line TTLCache with a lock, hit counter, and eviction policy — for a function that's called twice.

Snip shifts the default the other way.

snip: vs snip:prod — why this matters

No other tool in this space makes this distinction:

// snip: plain object dispatch; upgrade to class registry if handlers exceed 10
const handlers = { 'order.created': handleOrder, 'user.signup': handleSignup };
(handlers[event.type] ?? noop)(event);

// snip:prod — HMAC verification is a timing-attack boundary; do not simplify
const sig = Buffer.from(req.headers['x-signature'], 'hex');
const expected = createHmac('sha256', secret).update(req.rawBody).digest();
if (!timingSafeEqual(sig, expected)) return res.status(401).end();

Both are minimal. One is safe to shrink further. One is not. snip tells you which is which.

// snip:prod lines are excluded from the snip score — they are correctly complex by definition.

Install

Claude Code

/plugin install snip

Cursor

curl -o .cursor/rules/snip.mdc \
  https://raw.githubusercontent.com/amulyavarshney/snip/main/.cursor/rules/snip.mdc

Windsurf

curl -o .windsurf/rules/snip.md \
  https://raw.githubusercontent.com/amulyavarshney/snip/main/.windsurf/rules/snip.md

GitHub Copilot

curl -o .github/copilot-instructions.md \
  https://raw.githubusercontent.com/amulyavarshney/snip/main/.github/copilot-instructions.md

VS Code Extension

Search "Snip" in the VS Code Extension Marketplace, or:

code --install-extension amulyavarshney.snip

npm CLI

npm install -g @amulyavarshney/snip
snip init       # create .snip.json in current project
snip score src/ # score your codebase

Benchmark results

Measured across 6 production-realistic tasks (JWT auth middleware, Go rate limiter, webhook HMAC validation, config loader, retry logic, query builder). Median of 10 runs per cell.

| | baseline | snip full | snip ultra | |--|--|--|--| | LOC (median) | 68 | 14 | 8 | | Reduction | — | ~80% | ~88% |

Run yourself: npm run bench (requires ANTHROPIC_API_KEY)

Modes

| Mode | Behavior | |------|----------| | lite | Builds what's asked; names the leaner option in one line | | full | Full 7-rung ladder + language overlays (default) | | ultra | Deletion-first; challenges requirements before writing | | prod | Full ladder + // snip:prod protection for trust boundaries | | off | Deactivated |

/snip             activate at default
/snip ultra       switch to ultra
/snip prod        switch to prod (recommended for financial/auth codebases)
/snip lang go     switch language overlay
/snip-review      review current code for over-engineering

The 7-rung ladder

Before any code is written, snip stops at the first rung that holds:

0. Can this be deleted entirely?
1. Does this need to exist at all? (YAGNI)
2. Does the standard library do this? Use it.
3. Does a native platform feature cover it? Use it.
4. Does an already-installed dependency solve it? Use it.
5. Can it be one line? Make it one line.
6. Only then: write the minimum code that works.

Rung 0 is what makes snip different from other minimal-code tools — it fires on existing code, not just code being written.

Language overlays

Snip auto-detects your project's dominant language and adds idiomatic shortcuts.

Python — list comprehensions over loops, @dataclass, functools.lru_cache, pathlib.Path, contextlib.suppress

TypeScript — discriminated unions over class hierarchies, as const over enums, crypto.randomUUID() (Node 19+), fetch (Node 18+)

Go — errors.New + fmt.Errorf("%w") over custom error structs, table-driven tests, sync.Once, io.Reader/io.Writer

Disable per-language in .snip.json:

{ "overlays": { "python": false } }

Team config

Commit .snip.json to your repo. Every developer and CI run picks it up automatically.

{
  "mode": "full",
  "language": "auto",
  "prod": { "protect": ["auth", "payments", "billing"] },
  "score": { "minScore": 60, "failBelow": true }
}

CI gate

# .github/workflows/snip.yml
- name: Check snip score
  run: npx @amulyavarshney/snip score src/ --min-score 60 --fail-below

The snip score (0–100) measures how much of your codebase could be deleted or replaced with stdlib. snip:prod lines are excluded — they're correctly complex.

Before/after examples

| Example | Without | With | |---------|---------|------| | JWT auth middleware | 5 files, ~60 lines | 10 lines + snip:prod tag | | Retry with backoff | 3-class hierarchy, 120 lines | 12-line async loop | | Config loading | ConfigManager + 4 sources, 70 lines | 10 lines | | Webhook HMAC | 5 classes, 80 lines | 8 lines + snip:prod tag | | Go error types | 7 types, 50 lines | 4 sentinel vars |

Development

# Run all tests
npm test

# Sync IDE rule copies from rules/base.md
npm run sync

# Run benchmarks
npm run bench

Tests use Node.js built-in node:test. Zero test dependencies.

License

MIT