A curated collection of 11 Claude Code plugins and 7 marketplace skills, organized into a quality-gated marketplace that prioritizes depth over breadth. Each plugin is tested, security-reviewed, and documented to production standards rather than duplicating official Anthropic documentation.

Project Overview

Claude Code’s plugin system enables extending the AI coding assistant with custom commands, hooks, agents, and skills. This marketplace emerged from building real-world plugins for daily development work and recognizing the need for a curated, opinionated collection rather than a sprawling registry.

Architecture

Plugin Categories

The marketplace organizes plugins into five functional categories:

Integrations — Connect Claude Code to external services. Includes a personal CRM backed by Notion, and a biblical study tool using Strong’s Concordance.

Developer Tools — Curated best practices for specific frameworks. Vercel deployment skills and Prisma ORM patterns that go beyond generic documentation.

Workflow — Cross-session orchestration and retrospectives. The gastown-parallel-workflow plugin enables multi-agent coordination across git worktrees, while session-retro captures learnings at session boundaries.

Knowledge & Context — Context management infrastructure. The codified-context plugin implements a three-tier progressive disclosure system (constitution, trigger tables, subsystem maps) for managing large codebases.

Search & Discovery — The githits-mcp plugin provides semantic search across millions of open source code examples via MCP server integration.

Marketplace Skills

Seven reusable workflow skills available for installation:

• codification-loop — Capture debugging insights as durable context
• codified-context-patterns — Three-tier context framework
• context-health — Evaluate project context quality
• create-plugin — Scaffold new Claude Code plugins
• export-plan — Turn agent plans into shareable RFCs
• find-docs — Navigate plugin development documentation
• postmortem-writing — Structured postmortem authoring

Quality Gates

Every plugin passes through a review process before inclusion:

• Functional testing across Claude Code sessions
• Security review for hooks and bash execution
• Documentation standards with clear trigger descriptions
• No duplication of official Anthropic documentation
• Opinionated design — each plugin solves a specific workflow problem

Technical Implementation

Plugin Structure

Each plugin follows Claude Code’s plugin specification with plugin.json manifests, organized component directories for commands, skills, agents, and hooks, and MCP server configurations where needed.

Progressive Disclosure

Skills use YAML frontmatter with precise trigger descriptions so Claude Code loads them only when relevant, keeping the context window efficient. This pattern itself became a reusable skill (progressive-disclosure).

Git Hook Integration

Several plugins include hooks that fire on Claude Code events (PreToolUse, PostToolUse, Stop, SessionStart) enabling automated workflows like session retrospectives and code quality gates.

Design Philosophy

Quality over quantity — Better to have 11 well-tested plugins than 100 untested ones. Each plugin exists because it solved a real workflow problem.

Composable — Plugins work independently but compose naturally. Use codified-context with progressive-disclosure, or session-retro with beads task tracking.

Agent-native — Designed for AI agents as primary users. Clear trigger descriptions, structured outputs, and autonomous operation where appropriate.

Impact

• Reduced context-switching between tools during development sessions
• Established reusable patterns for plugin development adopted by other developers
• Demonstrated that plugin marketplaces benefit from curation and quality gates rather than open submission