Claude
Cursor
Skill
go-conventions
Go code conventions covering Effective Go, error wrapping, golangci-lint, govulncheck, table-driven tests, interfaces, context usage, modules, and concurrency. Load when writing or reviewing Go code.
Virus-scanned
Reviewed automatically before listing.
Download
goldziher-ai-rulez-internal_builtins_languages_go_skills_go-conventions-ca967ad.zip · 0 KB
Install
skills CLI
npx skills add https://github.com/Goldziher/ai-rulez/tree/main/internal/builtins/languages/go/skills/go-conventions
Claude Code
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install goldziher-ai-rulez@llmmart
Git
git clone https://github.com/Goldziher/ai-rulez.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole goldziher/ai-rulez collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
- Follow Effective Go and Go Code Review Comments guidelines.
- Handle every error return. Wrap errors with context:
fmt.Errorf("operation failed: %w", err). - Linting: a meta-linter (e.g., golangci-lint) with strict config (enable
govet,staticcheck,errcheck, and a security analyzer). Format withgofmt/goimports. - Security:
govulncheckfor CVE scanning, a SAST tool for static analysis. Run both in CI. - Testing: table-driven tests with
t.Run()and the stdlibtestingpackage. Uset.Parallel()where safe. An assertion library (e.g., testify) is optional. Coverage withgo test -coverprofile. - Naming: use short, descriptive names. Receivers are 1-2 letters. Exported names are descriptive.
- Prefer composition over inheritance. Use interfaces for abstraction (accept interfaces, return structs).
- Keep packages small and focused. Avoid package-level state and
init()functions. - Use
context.Contextas first parameter for cancelable operations. Never store contexts in structs. - Error types: use
errors.Is()/errors.As()for comparison. Define sentinel errors witherrors.New(). - Modules: use Go modules. Commit
go.sum. Use semantic version tags. Prefer stdlib over third-party when reasonable. - Concurrency: prefer channels over mutexes. Use
sync.WaitGroupfor fan-out. Guard shared state. - Benchmarking: use
testing.Bbenchmarks. Profile withgo tool pprof. Usebenchstatfor comparison.
Files (ai-rulez)
-
SKILL.md 1.6 KB
--- name: go-conventions description: Go code conventions covering Effective Go, error wrapping, golangci-lint, govulncheck, table-driven tests, interfaces, context usage, modules, and concurrency. Load when writing or reviewing Go code. --- - Follow Effective Go and Go Code Review Comments guidelines. - Handle every error return. Wrap errors with context: `fmt.Errorf("operation failed: %w", err)`. - Linting: a meta-linter (e.g., golangci-lint) with strict config (enable `govet`, `staticcheck`, `errcheck`, and a security analyzer). Format with `gofmt`/`goimports`. - Security: `govulncheck` for CVE scanning, a SAST tool for static analysis. Run both in CI. - Testing: table-driven tests with `t.Run()` and the stdlib `testing` package. Use `t.Parallel()` where safe. An assertion library (e.g., testify) is optional. Coverage with `go test -coverprofile`. - Naming: use short, descriptive names. Receivers are 1-2 letters. Exported names are descriptive. - Prefer composition over inheritance. Use interfaces for abstraction (accept interfaces, return structs). - Keep packages small and focused. Avoid package-level state and `init()` functions. - Use `context.Context` as first parameter for cancelable operations. Never store contexts in structs. - Error types: use `errors.Is()`/`errors.As()` for comparison. Define sentinel errors with `errors.New()`. - Modules: use Go modules. Commit `go.sum`. Use semantic version tags. Prefer stdlib over third-party when reasonable. - Concurrency: prefer channels over mutexes. Use `sync.WaitGroup` for fan-out. Guard shared state. - Benchmarking: use `testing.B` benchmarks. Profile with `go tool pprof`. Use `benchstat` for comparison.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.