tinymade-skills

A personal plugin marketplace for extending Claude Code with custom capabilities. Skills, commands, and agents—packaged and versioned for reuse across projects.

Problem

Claude Code is powerful out of the box, but specialized workflows need specialized tools. Rather than rebuilding the same integrations across projects, package them once and install everywhere.

Architecture

A monorepo of plugins with a key insight: separate skill definition from implementation.

Skill Definition (SKILL.md):

Human-readable markdown files
Describes purpose, usage patterns, decision logic
The agent “reads the manual” to learn the tool

Implementation (scripts):

Standard CLI tools (TypeScript/Bun, Python, Shell)
JSON output for reliable parsing
Stateless where possible

plugins/my-skill/
├── plugin.json         # Metadata
├── commands/           # Slash commands
│   └── my-command.md
├── agents/             # Specialized agents
│   └── my-agent.md
└── skills/             # Auto-triggering skills
    └── skill-name/
        ├── SKILL.md    # Definition with frontmatter
        ├── scripts/    # Implementation
        └── references/ # Reference docs

Available Plugins

Plugin	Purpose
gemini-offloader	Offload context-heavy tasks to Gemini with caching and mem0 memory
worktree-orchestrator	Git worktree management with terminal spawning
initializer	Parallel agent coordination with status TUI
plane	Plane.so integration—sync issues, link tasks

Key Pattern: Skill-as-Documentation

The most powerful concept is treating skills as documentation the agent reads, not code it executes blindly.

# When to Use This Skill

Use when the user asks to...
- Research topics requiring web grounding
- Summarize large documents (>50k tokens)
- Compare technologies

# Decision Logic

IF user query matches "research" pattern:
  → Use query.ts with --include-dirs
IF query mentions "session" or "continue":
  → Use session.ts continue

The agent reasons about when and how to use tools based on structured documentation, not hardcoded triggers.

Current Status

Why This Exists

AI agents need extensibility patterns that match how they think—through documentation and structured reasoning, not just function calls. This marketplace experiments with “skill-as-documentation” as a first-class concept for building agent capabilities.