Skip to content
/

sgaunet/claude-plugins

sgaunet-plugins · sgaunet/claude-plugins · ★ 13 · last commit 2026-05-07

Primitive shape 25 total
Commands 8 Skills 2 Subagents 14 MCP tools 1
00

Summary

sgaunet/claude-plugins — Summary

A curated four-plugin Claude Code marketplace by sgaunet (Sylvain Gaunet) targeting DevOps, infrastructure, software engineering, and Go development. The distinguishing architectural choice is a collection of deeply specialized domain-expert agents (14 total) rather than general-purpose skills — each agent is a named persona with strict role boundaries (e.g., aws-specialist, golang-pro, postgresql-specialist). The Go specialist plugin ships 6 specialized sub-agents including a go-blackbox (black-box test enforcement) and go-tdd agent, plus shell scripts for artifact generation (GitHub Actions, GitLab CI, GoReleaser). The repo optionally integrates with a perplexity-ai MCP server (built by the same author) for AI-powered web research within agent sessions, and references gh and glab CLIs as prerequisites for GitHub/GitLab operations. Total: 14 agents, 8 commands, 0 skills across 4 plugins.

differs_from_seeds: Closest to BMAD (Archetype 1: persona-md subagents) in organizing work through named specialist personas. However where BMAD personas are general software roles (architect, developer, QA), sgaunet's personas are infrastructure domain experts (AWS, PostgreSQL, Go, Terraform). No skills used — all work is agent-mediated. Unlike any seed, this marketplace explicitly integrates with external CLI tools (gh, glab) as hard prerequisites.

01

Overview

sgaunet/claude-plugins — Overview

Origin

Created and maintained by Sylvain Gaunet (sgaunet), a Go developer and DevOps practitioner. The repository has been iteratively developed to v0.18.3 on the go-specialist plugin, suggesting active ongoing development.

Philosophy

The README does not include a manifesto, but the design choices reveal the philosophy:

  • Domain specialists over generalists: Every agent is a named specialist with a defined scope
  • Production-ready tooling: References to real CLI tools (gh, glab, goreleaser, golangci-lint, biome) suggest practitioners building real systems
  • Proactive agents: The README states agents "automatically assist with their specialized domains" — they fire proactively rather than waiting for explicit invocation

From the code-review-enforcer agent prompt:

"Review every change like it's going to production tomorrow."

Agent Depth Philosophy

The code-review-enforcer agent is illustrative of the depth philosophy:

  • Four severity levels: Critical (blocks merge), High (should fix), Medium (improve), Low (consider)
  • Proactive triggers: "Automatically review when files modified via Edit/Write/MultiEdit operations"
  • Scoped tools: tools: Read, Grep, Glob, Bash(git:*) — explicit tool restriction
  • Explicit model assignment: model: sonnet

Plugin Versioning

The go-specialist plugin at v0.18.3 is the most mature. The semver versioning and CHANGELOG suggest sgaunet treats this as production software rather than a personal experiment.

02

Architecture

sgaunet/claude-plugins — Architecture

Distribution

  • GitHub: sgaunet/claude-plugins
  • Install: /plugin marketplace add sgaunet/claude-plugins then /plugin install <plugin>
  • Local dev: git clone + /plugin marketplace add ./claude-plugins
  • License: MIT
  • Primary language: Python (scripts) + Shell + Markdown (agents)

Directory Structure

claude-plugins/
├── .claude-plugin/           # Marketplace manifest
├── .claude/                  # Top-level Claude Code settings
├── .github/                  # CI workflows
├── .vscode/                  # VS Code config
├── CLAUDE.md                 # Project instructions
├── Taskfile.yml              # Task runner
├── extras/                   # Additional configs
├── requirements.txt          # Python dependencies
├── scripts/                  # Python helper scripts
└── plugins/
    ├── devops-infrastructure/
    │   ├── .claude-plugin/
    │   ├── agents/
    │   │   ├── aws-specialist.md
    │   │   ├── cicd-specialist.md
    │   │   ├── database-specialist.md
    │   │   ├── devops-specialist.md
    │   │   └── postgresql-specialist.md
    │   └── commands/
    │       └── analyze-db-performance.md
    ├── software-engineering/
    │   ├── .claude-plugin/
    │   │   ├── plugin.json
    │   │   └── .mcp.json        # Perplexity AI MCP config
    │   ├── agents/
    │   │   ├── code-review-enforcer.md
    │   │   ├── debugger.md
    │   │   ├── diagram-architect.md
    │   │   ├── docs-architect.md
    │   │   ├── html-first-frontend.md
    │   │   ├── license-specialist.md
    │   │   ├── payment-integrator.md
    │   │   └── security-auditor.md
    │   └── commands/
    │       └── (multiple commands)
    ├── go-specialist/
    │   ├── .claude-plugin/
    │   │   └── plugin.json       # v0.18.3
    │   ├── agents/
    │   │   ├── golang-pro.md
    │   │   ├── gen-github-dir.md
    │   │   ├── gen-gitlab-ci.md
    │   │   ├── gen-goreleaser.md
    │   │   ├── gen-linter.md
    │   │   ├── gen-taskfiles.md
    │   │   └── go-tdd.md
    │   └── commands/
    │       └── (multiple commands)
    └── bash-specialist/
        ├── agents/
        └── commands/

Optional MCP Integration

The software-engineering plugin includes .mcp.json referencing perplexity-ai MCP server (built by sgaunet). This adds AI-powered web search to agents in that plugin.

Required Runtime

  • Claude Code
  • gh CLI (GitHub operations)
  • glab CLI (GitLab operations)
  • Python 3 (for scripts)
  • perplexity-ai MCP server (optional, requires PPLX_API_KEY)
03

Components

sgaunet/claude-plugins — Components

devops-infrastructure Plugin

Agents (5):

Name Specialty
aws-specialist AWS cloud architecture, Well-Architected Framework, cost optimization
cicd-specialist GitHub Actions, GitLab CI, Forgejo Actions
database-specialist PostgreSQL, MySQL optimization and schema design
devops-specialist Terraform, Ansible, CloudFormation, cloud automation
postgresql-specialist PostgreSQL 16+ advanced features, query optimization, replication

Commands (1):

Name Purpose
analyze-db-performance PostgreSQL performance analysis with query and index insights

software-engineering Plugin

Agents (8):

Name Specialty
code-review-enforcer Code quality (Critical/High/Medium/Low severity), security, best practices. Proactive on file modifications. Tools: Read, Grep, Glob, Bash(git:*); model: sonnet
debugger Systematic debugging specialist
diagram-architect Architecture diagram generation
docs-architect Technical documentation specialist
html-first-frontend HTML-first frontend development (minimal JS philosophy)
license-specialist License compliance analysis
payment-integrator Payment system integration specialist
security-auditor Security vulnerability analysis

Commands: Multiple (exact count not enumerated from available data)

MCP: perplexity-ai (optional, for AI-powered research within agent sessions)

go-specialist Plugin (v0.18.3)

Agents (7):

Name Specialty
golang-pro Go 1.25+ expert — modern patterns, advanced concurrency, performance optimization
gen-github-dir Generate GitHub Actions CI workflows for Go
gen-gitlab-ci Generate GitLab CI pipelines for Go
gen-goreleaser Generate GoReleaser configs
gen-linter Generate golangci-lint configuration
gen-taskfiles Generate Taskfile.yml task runners
go-tdd Go test-driven development agent

Commands: Multiple (asset generation commands)

Skills (from plugin.json description):

  • go-blackbox — black box test enforcement
  • go-structure — project layout standards

bash-specialist Plugin

Agents: bash scripting specialist(s) Commands: bash-related commands

Total Primitive Count

  • Agents: 14+ (5 devops + 8 software-eng + 7 go-specialist + bash-specialist)
  • Commands: 8+ (1 devops + multiple others)
  • Skills: 2 (go-blackbox, go-structure — mentioned in plugin.json description)
  • Hooks: 0
  • MCP: 1 (optional perplexity-ai)
05

Prompts

sgaunet/claude-plugins — Prompts

Prompt File 1: code-review-enforcer agent (verbatim)

Technique: Severity-taxonomy enforcement with explicit proactive triggers and structured output format.

---
name: code-review-enforcer
description: Senior-level code reviewer. Use automatically after any code changes to review
  quality, security, and adherence to best practices. Should be invoked proactively
  when files are modified.
tools: Read, Grep, Glob, Bash(git:*)
model: sonnet
color: purple
---

You are a senior code reviewer. Review every change like it's going to production tomorrow.

## Proactive Triggers
Automatically review when:
- Files modified via Edit/Write/MultiEdit operations
- New files created in the codebase
- Before git commits (when requested)
- Pull request review needed

## Review Severity Levels

### 🔴 Critical (Must Fix - Blocks merge)
- **Security**: Exposed secrets/keys, SQL injection, XSS, auth bypass, CORS misconfiguration
- **Stability**: Crash conditions, data corruption, infinite loops, deadlocks, race conditions
- **Data Safety**: Unvalidated inputs, missing sanitization, PII exposure, missing encryption

### 🟡 High (Should Fix)
- **Logic Errors**: Off-by-one errors, incorrect conditionals, missing null checks
- **Error Handling**: Unhandled promises, missing try-catch for external calls
- **Performance**: Memory leaks, N+1 queries, inefficient algorithms (O(n²) where O(n) possible)
- **Breaking Changes**: API contract violations, schema changes without migration

### 🟢 Medium (Improve)
- **Code Quality**: Functions > 50 lines, duplicated code blocks, poor naming
- **Type Safety**: Missing/incorrect TypeScript types, any types, unchecked casts
- **Testing**: Missing tests for business logic, low coverage on critical paths

### ⚪ Low (Consider)
- **Documentation**: Unclear comments, outdated README
- **Optimization**: Missed caching opportunities
- **Style**: Formatting inconsistencies, unused imports

## Review Output Format

Code Review Summary

Files reviewed: X | Critical: X | High: X | Medium: X | Low: X

🔴 Critical Issues

  1. [SECURITY] SQL injection in users.go:45 Problem: Raw string concatenation in SQL query Fix: Use prepared statements or query builder

Prompting technique: Severity-graded taxonomy with explicit tool restriction (tools: Read, Grep, Glob, Bash(git:*)) and model assignment (model: sonnet). The 4-level severity taxonomy maps to merge-readiness decisions, making this an automated CI gate disguised as an agent.

Prompt File 2: go-specialist plugin.json metadata

Technique: Explicit keyword tagging for agent activation domain.

{
  "name": "go-specialist",
  "version": "0.18.3",
  "description": "Master Go 1.25+ development with modern patterns, advanced concurrency, performance optimization, and production-ready microservices. Skills: golangci-lint, goreleaser, go-tool (Go 1.24+ tool dependencies), go-blackbox (black box test enforcement), go-structure (project layout), GitHub Actions, GitLab CI. MCP: context7",
  "keywords": ["go", "golang", "goreleaser", "golangci-lint", "go-tool", "testing"],
  "author": {"name": "Sylvain", "email": "1552102+sgaunet@users.noreply.github.com"},
  "license": "MIT"
}

Prompting technique: Explicit capability claim in plugin description — keywords are not just discovery tags but communicate domain scope to the AI client loading the plugin.

09

Uniqueness

sgaunet/claude-plugins — Uniqueness

differs_from_seeds

Closest to BMAD (Archetype 1: persona-md subagents) in organizing work through named domain-specialist agents. However where BMAD personas are general software engineering roles (architect, developer, QA analyst), sgaunet's agents are infrastructure and language-domain specialists with explicit tool restrictions and model assignments. Unlike BMAD's single-plugin monolith (34 skills), sgaunet organizes into four independently installable domain plugins. The explicit tools: restriction in agent frontmatter (limiting code-review-enforcer to read-only tools) is a safety pattern not found in any of the 11 seeds. The optional perplexity-ai MCP integration for web research is a novel pattern for this batch.

Most Distinctive Feature

Per-agent tool restriction in YAML frontmatter (tools: Read, Grep, Glob, Bash(git:*)). This creates read-only, write-only, or restricted-write agents from a simple config field — a pragmatic approach to capability isolation without containers or worktrees.

Additionally, explicit model: sonnet assignment per agent enables role-specific model routing in a lightweight way — just a YAML field.

Positioning

  • DevOps/infrastructure-specialized (narrow domain, deep expertise per agent)
  • Production-oriented: "Review every change like it's going to production tomorrow"
  • Go-specialist at v0.18.3 is the most mature and versioned plugin
  • External CLI prerequisites (gh, glab) signal integration with real DevOps workflows

Observable Failure Modes

  1. CLI prerequisites: If gh or glab are not installed/authenticated, DevOps agents silently fail
  2. Perplexity API key required: The web-search integration requires a paid API key, creating a cost dependency
  3. No hooks: No lifecycle hooks means agents must be manually invoked (except for proactive trigger patterns in agent descriptions)
  4. No skills: All work is agent-mediated — there are no auto-activating skills for ambient guidance
  5. Go version drift: The go-specialist targets Go 1.25+; the plugin may become outdated as Go evolves rapidly
04

Workflow

sgaunet/claude-plugins — Workflow

Agent-Mediated Workflow

Work flows through named domain specialists. There is no explicit multi-phase workflow template; each agent handles its own domain end-to-end.

Code Review Workflow (code-review-enforcer)

File modified via Edit/Write/MultiEdit
  ↓ code-review-enforcer activates automatically
Review runs at four severity levels:
  Critical → Must Fix (blocks merge)
  High     → Should Fix
  Medium   → Improve
  Low      → Consider
Output:
  ## Code Review Summary
  Files reviewed: X | Critical: X | High: X | Medium: X | Low: X
  ### 🔴 Critical Issues
  ### 🟡 High Priority Issues
  ### ✅ Good Practices Observed

Go Development Workflow

User: "set up Go project with CI/CD"
  ↓ golang-pro agent handles overall development
  ↓ gen-github-dir agent generates GitHub Actions workflow
  ↓ gen-goreleaser agent generates release config
  ↓ gen-linter agent generates golangci-lint config
  ↓ go-tdd agent handles test-driven development

Phase-to-Artifact Map

Phase Artifact
DB performance analysis Performance report (analyze-db-performance command)
Go project setup GitHub Actions YAML, .goreleaser.yml, .golangci.yml, Taskfile.yml
Code review Severity-graded review report
Security audit Vulnerability report

Approval Gates

The code-review-enforcer uses severity levels as implicit gates (Critical = must fix before merge). No formal user-facing approval gate — the agent presents findings and the user decides.

Prerequisites Check

Agents that use gh or glab will fail if those CLIs are not authenticated. The README explicitly calls out:

"These plugins use gh and glab CLIs for GitHub/GitLab operations" "Authenticate after installation: gh auth login / glab auth login"

06

Memory Context

sgaunet/claude-plugins — Memory & Context

State Storage

No persistent state system. Agents work in-place; results are conversation output only.

Memory Type

None — file-based only in the sense that code changes persist to disk.

Cross-Session Handoff

None documented.

Context Injection

The software-engineering plugin includes a .mcp.json that adds the perplexity-ai MCP server to the agent's tool context. This extends the agent's information access (web search) rather than adding memory.

Agent Tool Restrictions

A key architectural choice in sgaunet-plugins is explicit tool restriction per agent. The code-review-enforcer restricts to tools: Read, Grep, Glob, Bash(git:*) — preventing accidental writes during review. This is a form of context/capability scoping rather than memory management.

Compaction Handling

Not documented. Agents are stateless; each invocation starts fresh.

07

Orchestration

sgaunet/claude-plugins — Orchestration

Multi-Agent

Yes — 14+ named agents. Each is independently invocable; no explicit orchestration layer.

Orchestration Pattern

Sequential (user activates one specialist at a time) or unorchestrated parallel (multiple agents available but no coordinator). No master orchestrator agent.

Subagent Definition Format

persona-md — each agent is an .md file with YAML frontmatter (name, description, tools, model, color) and role instructions.

Spawn Mechanism

Claude Code agent invocation system. Agents are invoked when their description matches the user's task, or when another agent delegates.

Isolation Mechanism

None. Work happens in-place.

Multi-Model

Partial. The code-review-enforcer agent specifies model: sonnet. Other agents likely default to the session model. This represents role-specific model assignment but only for one agent.

Execution Mode

Interactive-loop — agents fire based on task context.

Prompt Chaining

Yes — in the go development workflow, multiple generator agents (gen-github-dir, gen-goreleaser, etc.) chain together to produce a complete CI/CD setup.

Tool Restriction as Isolation

The code-review-enforcer uses tools: Read, Grep, Glob, Bash(git:*) to prevent write operations during review. This is a form of capability isolation even without environment isolation.

Optional Research Context

The perplexity-ai MCP server (software-engineering plugin) extends agents with web search capability — treating external research as a tool call rather than context injection.

08

Ui Cli Surface

sgaunet/claude-plugins — UI & CLI Surface

CLI Binary

None within this repo. The plugins use gh and glab CLIs as external prerequisites.

Local UI

None. No web dashboard or TUI.

IDE Integration

  • Claude Code: Primary target via /plugin marketplace system
  • Taskfile.yml: Developer task runner for maintaining the repo itself

External CLI Dependencies

Agents in this repo expect these CLIs to be available in the user's PATH:

  • gh (GitHub CLI) — for PR creation, issue management, GitHub Actions
  • glab (GitLab CLI) — for GitLab CI operations

This is a unique pattern in the corpus: the framework's agents have hard external CLI prerequisites that must be authenticated before agents can function.

Optional MCP Server

The software-engineering plugin can use perplexity-ai MCP server for AI-powered web research:

  • Install: go install from sgaunet/pplx repo
  • Configure: claude mcp add perplexity-ai -- pplx mcp-stdio
  • Required env: PPLX_API_KEY

Observability

The code-review-enforcer produces structured review reports (4-level severity breakdown) that serve as an audit trail for code changes.

Agent Color Coding

Agents specify a color field in YAML frontmatter (e.g., color: purple for code-review-enforcer). This surfaces in Claude Code's agent UI for visual identification.

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…