Once you start using Claude Code seriously, you accumulate prompts. A snippet that always produces the right golangci.yml. A spell that turns “I need a GitLab CI pipeline” into a working .gitlab-ci.yml. A code-review checklist you keep pasting in. After enough copy-paste, the obvious move is to package those into proper Claude Code plugins. That’s what claude-plugins is.
What’s in the marketplace
The repo is a Claude Code marketplace with three plugin collections, each focused on a domain I work in:
- devops-infrastructure — agents and commands for AWS, CI/CD, Terraform, Ansible, and PostgreSQL
- software-engineering — code review, debugging, docs, license compliance, security auditing, payment integration, plus a
/feature-flowcommand that handles the full git-branch-issue-commit dance - go-specialist — a Go 1.25+ expert agent and commands to scaffold
.golangci.yml, GitHub workflows, GitLab CI, GoReleaser configs, and Taskfiles
At the time of writing: 3 plugins, 14 agents, 22 commands. None of it invented out of nothing — each piece is a workflow I actually run.
Adding it to Claude Code
Once you have Claude Code installed, the marketplace is one command away:
/plugin marketplace add sgaunet/claude-plugins
Then list and pick what you want:
/plugin
For local hacking on the plugins themselves:
git clone git@github.com:sgaunet/claude-plugins.git
claude
# inside claude:
/plugin marketplace add ./claude-plugins
A few favorites
Some commands I lean on every day:
/feature-flow— given a description, it creates the branch, opens the issue, makes the commit, and keeps the metadata consistent across GitHub or GitLab./audit-codebase— a holistic security and performance pass instead of one-rule-at-a-time linting./gen-linter— emits a sane.golangci.ymlconfigured with 90+ linters, tuned for actual signal rather than noise./gen-gitlab-ciand/gen-github-dir— bootstrap a Go project’s CI from scratch with tests, coverage, and automated releases./gen-goreleaser— multi-arch Docker builds and a GoReleaser config that just works.
The golang-pro agent is the one I notice most, because it auto-activates on Go work and quietly steers things toward Go 1.25+ idioms (generics, structured concurrency, the new errors plumbing) without me having to ask.
Optional: Perplexity for research
For agents that need real-time information from the web, the marketplace plugs into pplx — my Perplexity AI MCP server — so research-heavy commands can get cited, current answers instead of the model’s recall:
claude mcp add perplexity-ai -- pplx mcp-stdio
export PPLX_API_KEY="your-key"
It’s optional. Without it, the agents still work, they just lose the live-web superpower.
The extras nobody asked for but I needed
Outside the plugin tree, the extras/ directory has two helpers I use on every Claude Code install:
statusline— a tiny bash statusline that shows the active model, a color-coded progress bar of the context window, and the total context size:[Opus] ▓▓▓░░░░░░░ 30% | 200k ctx. Glanceable and unobtrusive.no-leak— aPreToolUsehook that hard-blocks Claude Code from reading or modifying sensitive files (.env, credentials, private keys, vault files). The key word is hook — unlike CLAUDE.md instructions, hooks are deterministic and the LLM cannot ignore them. Belt and suspenders.
Both install with a single shell script.
Why publish this
Honestly? Mostly for me. Having my workflow as a versioned, validated marketplace means I can spin up a new machine, add the marketplace, and have my tooling back in seconds. But there’s a decent chance some of these commands save someone else an afternoon — and that’s a fine reason to publish.
Where to go next
- Source: github.com/sgaunet/claude-plugins
- Plugin development docs: docs.claude.ai/claude-code/plugins
It’s open source, MIT-licensed, and PRs are welcome — especially for new specialized agents or commands you’d like to see. If you’re a Claude Code user who lives in the DevOps or Go world, the cost to try is exactly one slash command.