cc-skills-golang

cc-skills-golang is an AI Agent Skills collection for production-grade Go projects, maintained by samber (Samuel Berthe), currently at v1.4.0. The core idea is to decompose Go engineering best practices into atomic, cross-referenced Markdown instruction sets that are loaded on-demand into various AI coding assistants.

Skill System#

Skills are divided into two categories:

General-purpose skills (29): Covering code style, data structures, databases, design patterns, documentation, error handling, modernization, naming conventions, security, testing, performance, concurrency, context management, CI/CD, dependency injection, dependency management, interfaces & structs, linting, observability, project layout, etc. 13 are recommended.

Tool skills (14): Covering Google Wire, gRPC, GraphQL, Cobra CLI, Viper configuration, Swagger, Uber Dig/Fx, samber ecosystem libraries (do, hot, lo, mo, oops, ro, slog), stretchr/testify, etc.

Scheduling Mechanisms#

• On-demand loading: Description field (~100 tokens) always loaded for trigger detection; full SKILL.md loaded only when activated; secondary markdown files lazy-loaded to preserve context budget
• Atomic + cross-referencing: Each concept belongs to exactly one skill; other skills reference via samber/cc-skills-golang@<skill> format to avoid duplication
• Override mechanism: Enterprises can declare overrides for community default skills in their own skill definitions
• Command triggering: Some skills support slash command manual invocation (e.g., /golang-performance)
• Deep thinking: Security auditing, performance analysis, troubleshooting skills are tagged for ultrathink mode

Token Budget Control#

• Single SKILL.md recommended < 2,500 tokens
• Total simultaneous skill loading recommended < 10,000 tokens

AI-Driven Code Review (CI Integration)#

Complete GitHub Actions workflows are provided to use AI Agents as PR Reviewers:

Evaluation Data#

The project's adversarial evaluation shows 98% accuracy with skills (3315/3395) vs 56% without (1915/3395), a +41pp improvement. This data is based on the project's own evaluation set; per-skill breakdown details are not fully publicly available.

Installation & Usage#

Unified installation via npx skills CLI:

# Install all skills
npx skills add https://github.com/samber/cc-skills-golang --all

# Install single skill
npx skills add https://github.com/samber/cc-skills-golang --skill golang-performance

# CI integration
npx skills add https://github.com/samber/cc-skills-golang -a claude-code --skill '*' -y --copy

Supported parameters: --all (install all), --skill <name> (specific skill, supports '*' wildcard), -a <agent> (target agent), --copy (copy to local .agents/ directory), -y (skip confirmation). Plugin configs .claude-plugin/plugin.json, .cursor-plugin/plugin.json, and gemini-extension.json must maintain version consistency.

Compatibility#

Compatible with Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Codex, Antigravity, Openclaw and other mainstream Agent tools, while also compatible with Snyk agent scanner security rules (W011/W012/W001). GitHub Copilot integration is via .github/copilot-instructions.md in the project root.

Architecture Overview#

skills/
  <skill-name>/
    SKILL.md          # Required: YAML frontmatter + instruction body
    references/       # Optional: on-demand detailed docs
    scripts/          # Optional: executable scripts
    assets/           # Optional: templates, resources, linter configs, etc.
    evals/            # Optional: adversarial evaluation scenarios

Follows the Agent Skills Specification, integrated with the ClawHub skill marketplace.