Skip to content

Home

Jump to bottom
Chris Purcell edited this page Feb 20, 2026 · 3 revisions

This page is auto-generated from the repository README. Do not edit directly — changes will be overwritten on the next sync.

Claude Code Plugins Marketplace

A Claude Code plugin marketplace covering orchestration, release automation, design review, Home Assistant development, Linux system administration, GitHub repository management, and plugin testing.

Installation

Add this marketplace to your Claude Code installation:

/plugin marketplace add L3DigitalNet/Claude-Code-Plugins

Or using the full URL:

/plugin marketplace add https://github.com/L3DigitalNet/Claude-Code-Plugins.git

Then install individual plugins:

/plugin install agent-orchestrator@l3digitalnet-plugins

Staying Up to Date

Auto-update keeps plugins current automatically. To verify it's enabled:

  1. Run /plugin in Claude Code
  2. Go to the Marketplaces tab
  3. Select l3digitalnet-plugins
  4. Look for Disable auto-update (meaning it's already on)

When auto-update is enabled, Claude Code refreshes the marketplace catalog and updates installed plugins at the start of each session.

Manual update if you prefer to control when updates happen:

# Refresh the marketplace catalog
/plugin marketplace update l3digitalnet-plugins

# Then update individual plugins via /plugin → Installed tab → Update now

Available Plugins

Plugin Type Command Description
Agent Orchestrator Commands + Hooks /orchestrate Delegates complex tasks to agent teams with context management
Design Assistant Commands + Skills /design-draft, /design-review Guided design document authoring and principle-enforced review
GitHub Repo Manager Commands + Skills /repo-manager Conversational GitHub repo health assessment and maintenance
Home Assistant Dev Commands + Skills + MCP varies Full HA integration development toolkit with 27 skills
Linux SysAdmin MCP MCP ~100 tools Linux system administration across 15 modules
Plugin Test Harness MCP 18 tools Iterative test/fix/reload loop for plugin development
Release Pipeline Commands + Skills /release Semver releases with pre-flight checks and changelog generation

Principles

These principles apply across all plugins in this collection. Individual plugins may define additional principles scoped to their domain.

[P1] Composable Over Monolithic — Functionality is partitioned into focused, independently useful units. Complex workflows emerge from combining atomic components at runtime — not from monolithic mega-commands or mega-documents. Each unit does one thing well; orchestration is assembled from the outside, not baked in.

[P2] Scope Fidelity — A plugin does exactly what it was invoked to do — no more. Observations outside the declared scope are surfaced as notes, never acted upon. Mutations outside the session's stated purpose require explicit re-invocation. Scope creep is a trust violation.

[P3] Safe Default, Explicit Escalation — The default posture of every operation is the minimal, reversible, lowest-impact option available. Destructive modes, broad scopes, and irreversible paths are never the default; the user must take an explicit step to reach them. Dry-run before live, read before write, preview before execute, narrow scope before wide.

[P4] Human Consent Before Consequence — No plugin autonomously performs a destructive or irreversible action. Every high-impact operation is announced with enough context to make an informed decision, and irreversible actions are flagged unmissably before execution. Silent mutations — deployments, deletions, pushes, or state changes — are not possible. Confirming once in one context does not authorise the same action in another.

[P5] Explainability Precedes Action — Before a plugin acts, it explains what it is about to do and why — in plain language, with technical jargon translated on first mention and consequences surfaced before execution. A user's understanding of an operation is a prerequisite for meaningful consent, not a bonus. Explanations are proportionate: brief for routine actions, detailed for high-impact ones.

[P6] Conversation-Native Interaction — Plugins are experienced through Claude Code's terminal conversation, and every UX decision must honour that medium. Prefer bounded AskUserQuestion choices over open-ended text prompts. Lead with the answer — most important information first, context after. Use progressive disclosure over walls of text; any output longer than ~10 lines should be restructured or chunked. Format purposefully (headers, bold, status symbols) but never decoratively. Smart defaults eliminate questions the user shouldn't need to answer. Errors are actionable: say what went wrong and what to do about it — never a raw stack trace.

[P7] Fail Fast, Never Silently — When something goes wrong, stop immediately and surface the complete failure. Raw output is always shown alongside any interpretation; errors are never swallowed or softened into ambiguity. No autonomous recovery attempts, no continuing past a known failure. The user should never wonder whether an operation succeeded.

[P8] Done is Measured, Not Declared — Plugins that perform iterative work define completion as a measurable state — zero findings, all checks green, all tests passing — and drive toward that state across successive cycles. Success is not declared after a single pass; the plugin reports the trend and continues until the criterion is met or the user stops it.

[P9] One Concept Per Skill — Every skill covers a single, tightly-scoped concept — narrow enough that its trigger description fits in one sentence without "or". Skills load in full when triggered; a skill spanning five related topics loads all five even when only one applies. Wide-scope skills silently tax every loosely related task with tokens that do no work. If a skill's trigger description requires "or" to cover its own content, it's two skills.

Agent Orchestrator

General-purpose agent team orchestration with automatic context management, file isolation via git worktrees, and mechanical enforcement hooks.

Features:

  • Triage gate for simple vs complex tasks
  • Parallel execution via agent teams (or sequential fallback)
  • Git worktree isolation for concurrent work
  • Context degradation prevention via hooks
  • Quality gate with integration checking

Install:

/plugin install agent-orchestrator@l3digitalnet-plugins

Learn more: plugins/agent-orchestrator/README.md

Design Assistant

Full design document lifecycle in two commands — guided authoring from blank page through principle-enforced iterative review.

Features:

  • /design-draft — 5-phase interview: context deep dive, principles discovery, scope confirmation, gap-filling, and draft generation
  • Principle stress-testing and tension resolution before any architecture is committed
  • /design-review — multi-pass principle enforcement, gap analysis, and optional auto-fix
  • Principle Conflict Screening: all proposed fixes checked against established principles before presentation
  • Automatic warm handoff from draft to review (principles registry transferred)
  • Runs until the document converges to zero findings across all review tracks

Install:

/plugin install design-assistant@l3digitalnet-plugins

Learn more: plugins/design-assistant/README.md

GitHub Repo Manager

Conversational GitHub repository maintenance — assess and fix repo health interactively, with owner approval at every step.

Features:

  • Wiki sync — keeps wiki pages in sync with in-repo docs and code
  • Community health — audits and updates CONTRIBUTING, SECURITY, CODE_OF_CONDUCT, issue/PR templates
  • PR triage — conflict detection, staleness checks, review summaries
  • Issue triage — labels, assignees, staleness, linked PRs
  • Release health — unreleased commits, changelog drift, draft releases
  • Security posture — Dependabot alerts, code scanning, secret scanning
  • Dependency audit — outdated packages, license concerns
  • Notifications & discussions — triage and summarise
  • Cross-repo mode — scan all accessible repos for a specific concern and fix in batch
  • Structured maintenance report generated at session end

Install:

/plugin install github-repo-manager@l3digitalnet-plugins

Learn more: plugins/github-repo-manager/README.md

Home Assistant Dev

Comprehensive Home Assistant integration development toolkit with 19 AI skills, an MCP server for live HA connections, automated validation, example integrations, and project templates.

Features:

  • 27 context-aware skills covering architecture, config flows, coordinators, entities, testing, and more
  • 3 specialized agents (development, review, debugging)
  • MCP server with 12 tools for live HA connection and documentation search
  • 5 validation scripts with PostToolUse hook enforcement
  • 3 example integrations (Bronze/Silver/Gold tier)
  • 9 project templates for CI/CD, testing, and documentation
  • Full Integration Quality Scale coverage (all 52 rules)

Install:

/plugin install home-assistant-dev@l3digitalnet-plugins

Learn more: plugins/home-assistant-dev/README.md

Linux SysAdmin MCP

Comprehensive Linux system administration MCP server with ~100 tools across 15 modules for managing packages, services, users, firewall, networking, security, storage, containers, and more.

Features:

  • ~100 tools organized across 15 modules (packages, services, users, firewall, networking, security, storage, performance, logs, containers, SSH, cron, backup, docs, session)
  • Distro-agnostic command abstraction (Debian/RHEL auto-detection)
  • 5-tier risk classification with confirmation gates
  • YAML knowledge profiles for 8 services (sshd, nginx, docker, ufw, fail2ban, pihole, unbound, crowdsec)
  • SSH remote execution support
  • Git-backed documentation generation

Install:

/plugin install linux-sysadmin-mcp@l3digitalnet-plugins

Learn more: plugins/linux-sysadmin-mcp/README.md

Plugin Test Harness

Iterative plugin testing framework — generates tests, records pass/fail results, applies source fixes, reloads the target plugin, and retests until convergence.

Features:

  • Auto-generates tests from plugin source and schema introspection
  • Test/fix/reload loop with convergence tracking (improving, plateau, oscillating, diverging)
  • Dedicated git branch per session for a complete audit trail of fixes
  • Sessions persist to disk and can be resumed after interruption
  • Native MCP client for MCP-mode plugins; source analysis for hook/command plugins
  • 18 tools across session management, test management, execution, and fix management

Install:

/plugin install plugin-test-harness@l3digitalnet-plugins

Learn more: plugins/plugin-test-harness/README.md

Release Pipeline

Autonomous release pipeline — quick merge or full semver release with parallel pre-flight checks, changelog generation, and GitHub release creation.

Features:

  • Two modes: quick merge (testing → main) or full versioned release
  • Parallel pre-flight agents (test runner, docs auditor, git validator)
  • Automatic changelog generation from conventional commits
  • Version bumping across Python, Node.js, Rust, and plugin manifests
  • GitHub release creation with release notes
  • Human-in-the-loop approval gates at critical stages
  • Fail-fast with rollback guidance on errors

Install:

/plugin install release-pipeline@l3digitalnet-plugins

Learn more: plugins/release-pipeline/README.md

Coming Soon

These plugins are in development and not yet available in the marketplace.

Plugin Description
performance-profiler Latency measurement, flamegraph generation, and regression tracking for MCP servers
docs-manager Documentation lifecycle management — audit, index, and organise project docs

Plugin Development

This repository also serves as a development workspace for creating new plugins. See CLAUDE.md for architectural guidance and docs/ for comprehensive documentation.

Quick Start

  1. Create a new plugin:

    mkdir -p plugins/my-plugin/.claude-plugin
cd plugins/my-plugin

  2. Add manifest:

    {
  "name": "my-plugin",
  "version": "0.1.0",
  "description": "Plugin description"
}

  3. Test locally:

    claude --plugin-dir ./plugins/my-plugin

  4. Add to marketplace catalog (.claude-plugin/marketplace.json)

Documentation

Repository Structure

Claude-Code-Plugins/
├── .claude-plugin/
│   └── marketplace.json        # Marketplace catalog
├── plugins/                     # All plugin implementations
│   ├── agent-orchestrator/      # Agent team orchestration
│   ├── design-assistant/        # Design document lifecycle
│   ├── docs-manager/            # Documentation management (in development)
│   ├── github-repo-manager/     # Conversational GitHub repo maintenance
│   ├── home-assistant-dev/      # Home Assistant integration dev toolkit
│   ├── linux-sysadmin-mcp/      # Linux sysadmin MCP server (~100 tools)
│   ├── performance-profiler/    # MCP performance profiling (in development)
│   ├── plugin-test-harness/     # Iterative plugin testing framework
│   └── release-pipeline/        # Autonomous release pipeline
├── scripts/
│   └── validate-marketplace.sh  # Marketplace validation
├── docs/                        # Comprehensive documentation
├── CLAUDE.md                    # Development guidance for AI agents
├── BRANCH_PROTECTION.md         # Branch protection and workflow guide
└── README.md                    # This file

Contributing

To add a plugin to this marketplace:

  1. Work on the testing branch (all development happens here)
  2. Create plugin in plugins/ directory
  3. Add entry to .claude-plugin/marketplace.json
  4. Validate with ./scripts/validate-marketplace.sh
  5. Commit and push to testing branch
  6. When ready to deploy, merge testing → main

Branch workflow:

  • main - Protected production branch (GitHub blocks direct pushes)
  • testing - Development branch (direct commits allowed)

Deployment:

git checkout testing
./scripts/validate-marketplace.sh
git checkout main
git merge testing --no-ff -m "Deploy: <description>"
git push origin main
git checkout testing

See BRANCH_PROTECTION.md for detailed workflow documentation.

License

MIT - See LICENSE file for details

Last synced: 2026-02-20 · View source

Claude Code Plugins

Plugins

View source

Clone this wiki locally