Skip to content

Claude Code skills, agents and plugins

The tools I build and run for agentic software engineering, where AI agents write code under human approval.


The full set

Each entry is something I run on real work, and every one links to a public repo you can install from today.

SpecForge

Spec-driven AI development lifecycle with hook-enforced approval gates

Skill

SpecForge keeps Claude Code on a spec and plan you approved. It merges GitHub Spec Kit with AWS AI-DLC.

  • Runs seven stages, from constitution through plan to implement
  • A PreToolUse hook blocks code edits until a human approves the spec and plan
  • Reverse-engineers existing codebases into specs before any change
  • Installs as a Claude Code plugin or as agent skills via skills.sh
Claude CodeSpec-DrivenAI-DLCAgent SkillsGovernance
$ npx skills add pooyagolchian/specforge

use-local-llm

React hooks for streaming local LLM responses

Plugin

Stream replies from a model on your own machine into a React app.

  • Streams from Ollama, LM Studio, llama.cpp, and OpenAI-compatible endpoints
  • Runs direct browser to localhost with no server or API routes
  • Manages message history, abort, clear, and error handling
  • Ships at 2.8 KB gzipped with zero runtime deps beyond React
ReactTypeScriptLLMOllamaStreaming
$ pnpm add use-local-llm

RevOps AI

AI sales agent built on Notion through the Model Context Protocol

Agent

Gemini 2.5 Flash works your Notion sales pipeline, from pre-call briefings to deal updates.

  • Chains 22 Notion MCP operations per request autonomously
  • Scores leads 0–100 with written reasoning, saved back to Notion
  • Syncs a drag-and-drop Kanban pipeline to Notion in real time
  • Answers pipeline queries and drafts emails in natural language
AINotionMCPGeminiRevOps