Generalv1.2.1

claude-extensibility

Tools and skills for extending Claude Code with custom skills, agents, commands, and MCP servers

by DX TeamView on GitHub ↗
skillsagentsextensibilitydevelopmentplugin-developmentclaude-codeauditbloatreview

Claude Extensibility

Tools and skills for extending Claude Code with custom skills, agents, commands, and MCP servers.

Installation

claude plugin install --marketplace https://github.com/Personio-Internal/claude-code-marketplace --plugin claude-extensibility

What does this plugin do?

This plugin helps you create new capabilities for Claude Code. It provides structured guidance, scripts, and best practices for building skills that extend Claude with specialized knowledge, workflows, and tool integrations.

/skill-creator

A comprehensive guide for creating effective Claude Code skills from scratch or improving existing ones.

What you need:

  • Node.js (for the bundled helper scripts)
  • A clear idea of what capability you want to create

Quick Start:

# Tell Claude what you want to build
"Create a skill that helps me generate database migration scripts"

Claude will walk you through the full skill creation process: understanding requirements, planning resources, initializing the skill structure, implementing it, and packaging it for distribution.

Example:

Say you want to create a skill that helps generate API documentation from code:

You: "I want to create a skill that generates OpenAPI specs from my Express routes"

Claude will:
1. Ask clarifying questions about your routes and documentation needs
2. Plan what scripts, references, and assets the skill needs
3. Initialize a new skill directory with proper structure
4. Help you implement the SKILL.md and any bundled resources
5. Validate and package the skill as a distributable .skill file

The skill creator includes two helper scripts:

  • init_skill.js - Scaffolds a new skill directory with templates and proper structure
  • quick_validate.js - Validates skill structure and SKILL.md frontmatter

It also includes reference guides on proven design patterns:

  • Output patterns - Templates and examples for consistent skill output
  • Workflow patterns - Sequential and conditional workflow design

Why would you want this?

  • Consistent structure: Every skill follows the same proven conventions, making them easier to maintain and share
  • Guided process: Step-by-step workflow ensures you don't miss important details like proper frontmatter, resource organization, or progressive disclosure
  • Bundled tooling: Init, validate, and package scripts eliminate manual setup and catch errors early
  • Best practices built in: Incorporates lessons from Anthropic's official skill design patterns

trim-skill

Audit an existing SKILL.md for bloat and propose specific cuts with line references. Pairs with skill-creator — use it after a skill has been created and has grown over time (often via LLM-assisted edits that accumulate quietly).

What you need:

  • A SKILL.md file (or a skill directory) you want to review

Quick Start:

"Trim plugins/foo/skills/bar/SKILL.md"
"Audit this SKILL.md for bloat"
"Review this skill before I merge the PR"

Example:

You: "Trim plugins/pr-workflow/skills/ship/SKILL.md"

Claude will:
1. Grep for red-flag section headers (Notes for the AI Agent, Success Criteria, Tips, Best Practices, etc.)
2. Walk the file against the bloat pattern list (TAIL-META, STEP-BLOAT, OVERVIEW-DUP, GENERIC-METHODOLOGY, DEFENSIVE-ERRORS, OVER-EXAMPLES, HANDWAVY, KEY-POINTS, NARRATIVE-BLOAT, COSMETIC-OUTPUT, PREREQUISITE-AS-PHASE)
3. Report a cut list with line references, estimated bloat %, and one-line rationale per cut
4. Ask which cuts to apply before editing — defaults to keeping when unclear

The skill leans on Anthropic's skill authoring best practices for the upstream baseline (500-line ceiling, frontmatter limits, progressive disclosure) and focuses on the recurring semantic patterns a real marketplace accumulates over time. It does not touch CLAUDE.md — use the claude-md skill from personio-common for that.

Why would you want this?

  • Keeps skills reviewable: Prevents Claude-generated bloat from quietly accumulating across iterations — reviewers disengage when diffs are large, which is exactly how the drift compounds
  • Cuts token cost: Every skill loads its body into context when triggered; tighter skills mean better attention on every request
  • Catches the delta from upstream: Focuses on so-far identified patterns (TAIL-META tail sections, OVERVIEW-DUP, generic-methodology skills) that generic linters might not flag
  • Defaults to safe: Proposes cuts before applying them — false-positive cuts are worse than leaving bloat in