claude-skills

Twelve Claude Code skills that enforce strict code quality, security, and modern language patterns. Drop them into ~/.claude/skills/ and they trigger automatically when you work in the matching language.

What's inside

Skill	Scope
typescript-strict	TS 6.0 / 7.0 baseline, no any, no as, exhaustive unions, using, satisfies, verbatimModuleSyntax
rust-strict	Edition 2024 rules, unwrap ban, error hierarchy, unsafe audit, newtype IDs, tracing
swift-strict	Swift 6.2, strict concurrency, typed throws, Sendable, privacy manifest, InlineArray
go-strict	Go 1.26, log/slog, context propagation, errors.Is/As, os.Root, range-over-func
javascript-strict	Node 24 LTS, AbortController, crypto.randomBytes, streams, stdlib over deps
postgres-strict	PostgreSQL 18, async I/O, migration safety (CONCURRENTLY, NOT VALID), RLS, pgvector HNSW
security-audit-standard	OWASP Top 10, secret scan, supply chain, auth/authz, audit report format
performance-audit-standard	Big-O hot path audit, N+1 queries, indexes, Core Web Vitals targets
testing-patterns	Vitest, cargo test, Swift Testing, go test, property-based, coverage targets
code-review	Multi-language review pipeline that calls the strict rules above
github-standards	PR template, branch naming, CODEOWNERS, action SHA pinning, secret scanning
git-commit	Conventional Commits with the "why" question and storytelling body

Each skill is one SKILL.md file. Read-only standards, no scripts.

Install

# Clone into your skills directory
git clone git@github.com:0xMassi/claude-skills.git ~/Developer/claude-skills
cd ~/Developer/claude-skills

# Symlink each skill into the active skills folder
for s in */SKILL.md; do
  ln -s "$PWD/$(dirname $s)" ~/.claude/skills/
done

# Optional: pair with the included global CLAUDE.md
cp CLAUDE.md ~/.claude/CLAUDE.md

Restart Claude Code. The skills surface in the skills list and load when their description matches the task.

Pair with CLAUDE.md

The repo also ships a global CLAUDE.md that sets the high-level posture (senior-dev mindset, YAGNI, goal-driven on non-trivial work, direct communication). The skills enforce what the code should look like, the CLAUDE.md sets how the assistant behaves. Use both or use neither.

How they work

Skills are markdown files with a frontmatter description. Claude Code reads the description on every turn and pulls the body in only when relevant. So rust-strict loads when you touch .rs files, postgres-strict loads when you write a migration, code-review loads when you say /code-review.

There's no hook, no shell wrapper, nothing background. The whole system is "the right rules end up in context at the right time."

Versions targeted

Tool	Stable
TypeScript	6.0 (7.0 Beta available)
Rust	1.95.0, edition 2024
Swift	6.2
Go	1.26
Node.js	24 LTS Krypton
PostgreSQL	18

Rules cite specific compiler flags, stdlib packages, and SQL features that exist in these versions. Older codebases still benefit from the structural rules but version-specific items will not apply.

What this is not

• Not a linter. Linters check syntax, these check intent.
• Not a tutorial. Each rule assumes you already know the language.
• Not exhaustive. Twelve files, ~4700 lines. Strict means opinionated, not complete.

Conventions inside each file

• BAD / GOOD code blocks side by side
• Severity tags: CRITICAL / HIGH / MEDIUM / LOW
• Numbered rules (TS-01, GO-22, PG-15) you can reference in PR comments
• A vulnerability or audit checklist at the bottom

License

MIT. Use them, fork them, ship them.