SkillForge (skf)

The Complete AI Development Toolkit for Claude Code

92 skills | 10 categories | 20 agents | 90 hooks (24 registered) | 4 tiers

---

Transform Claude Code into a full-stack AI development powerhouse. From RAG pipelines to React 19 patterns, from database schemas to security audits - everything you need to build production-grade applications with AI assistance.

---

About SkillForge

SkillForge transforms Claude Code into a comprehensive AI development platform by providing:

• 92 Skills: Reusable knowledge modules organized in 10 categories that load on-demand
• 20 Agents: Specialized AI personas with domain expertise and pre-loaded skills
• 90 Hooks: Lifecycle automation for safety, auditing, and quality gates (24 registered via dispatchers)
• Progressive Loading: Token-efficient system that loads only what's needed (~300-800 tokens vs 5000+)

Why SkillForge?

Without SkillForge	With SkillForge
Manual context management	Automatic progressive loading
Generic AI responses	Domain-expert agents (FastAPI, React 19, LangGraph)
No guardrails	90 hooks for safety, auditing, quality gates
Single-threaded	Parallel agent execution (fan-out/fan-in)
Context overload	Token-efficient (~70% savings)

What's New in v4.11.0 (Hook Consolidation)

• Hook Consolidation: Reduced from 44 to 23 registered hooks using dispatcher pattern (48% reduction)
• MCP Updates: Added mem0 (cloud semantic memory) alongside Anthropic memory
• Fixed Paths: All hook references now correctly point to existing files
• New Dispatchers: agent-dispatcher, skill-dispatcher, session-end-dispatcher
• Cleaned Dead Code: Removed 9 unused hook files

Previous (v4.10.0 - CC 2.1.6 Integration)

skills/
├── ai-llm/.claude/skills/       # 19 skills: RAG, embeddings, agents, caching
├── langgraph/.claude/skills/    # 7 skills: State, routing, parallel, checkpoints
├── backend/.claude/skills/      # 15 skills: FastAPI, architecture, databases
├── frontend/.claude/skills/     # 6 skills: React 19, design systems
├── testing/.claude/skills/      # 9 skills: Unit, integration, E2E, mocking
├── security/.claude/skills/     # 5 skills: OWASP, auth, validation
├── devops/.claude/skills/       # 4 skills: CI/CD, observability
├── workflows/.claude/skills/    # 13 skills: Git, PR, implementation
├── quality/.claude/skills/      # 8 skills: Quality gates, reviews
└── context/.claude/skills/      # 6 skills: Compression, brainstorming

Quick Start

From Marketplace (Pending Approval)

Status: Awaiting approval at Anthropic Official and CC Marketplace. Once merged, these commands will work:

# Step 1: Add the marketplace
/plugin marketplace add yonatangross/skillforge-claude-plugin

# Step 2: Install
/plugin install skf

From GitHub (Manual)

git clone https://github.com/yonatangross/skillforge-claude-plugin ~/.claude/plugins/skillforge

Project-Scoped (Copy to Project)

# Copy entire plugin
cp -r skillforge-claude-plugin/.claude your-project/.claude

# Or copy specific skill categories (CC 2.1.6 auto-discovers them!)
cp -r skillforge-claude-plugin/skills/ai-llm your-project/

Configuration

Interactive Wizard (Recommended)

/skf:configure

The wizard guides you through:

1. Preset Selection: Choose your starting point

   • complete - All 92 skills, 20 agents, full hooks (default)
   • standard - All skills, no auto-agents, full hooks
   • lite - Essential skills only, minimal context
   • hooks-only - Safety guardrails only

2. Skill Categories: Toggle categories on/off based on your stack

3. Agent Configuration: Enable/disable specific agents

4. Hook Settings: Customize safety and quality hooks

5. MCP Setup: Enable optional MCP integrations

Configuration File

Settings stored in ~/.claude/plugins/skillforge/config.json:

{
  "preset": "complete",
  "skills": {
    "ai-llm": true,
    "langgraph": true,
    "backend": true,
    "frontend": true,
    "testing": true,
    "security": true,
    "devops": true,
    "workflows": true,
    "quality": true,
    "context": true
  },
  "agents": { "enabled": true, "allowed": ["*"] },
  "hooks": { "git_branch_protection": true, "file_guard": true }
}

Per-Project Overrides

Create .claude/skillforge.json in your project root:

{
  "skills": { "langgraph": false },
  "agents": { "allowed": ["backend-system-architect", "database-engineer"] }
}

---

Claude Code Integration (CC 2.1.6+)

SkillForge leverages Claude Code 2.1.6+ features for optimal performance.

Spawning Agents

Agents are spawned using the Task tool with subagent_type:

# Spawn a single agent
Task(
    subagent_type="skf:backend-system-architect",
    prompt="Design the user authentication API"
)

Parallel Execution (Fan-Out)

Launch multiple agents simultaneously by making multiple tool calls in a single message:

# These run concurrently - 3x faster than sequential
Task(subagent_type="skf:frontend-ui-developer", prompt="Build dashboard UI")
Task(subagent_type="skf:backend-system-architect", prompt="Build dashboard API")  
Task(subagent_type="skf:database-engineer", prompt="Design dashboard schema")

Context Modes

Control how agents share context via frontmatter:

Mode	Behavior	Use Case
fork	Isolated context (default)	Complex multi-step operations
inherit	Share parent context	Quick utilities, saves tokens
none	No context management	Stateless coordination

---
name: my-agent
context: fork  # isolated context
---

Model Selection

Agents specify preferred models for cost/performance tradeoffs:

Model	Best For	Cost
opus	Complex reasoning, architecture design	$$$
sonnet	Balanced tasks, implementation (default)	$$
haiku	Fast routing, simple validation	$

---
name: security-auditor
model: haiku  # Fast scanning, low cost
---

Skills Auto-Injection

CC 2.1.6 automatically injects skills listed in agent frontmatter:

---
name: backend-system-architect
skills:
  - api-design-framework
  - clean-architecture
  - fastapi-advanced
---

When this agent spawns, all three skills are automatically available - no manual loading required.

---

Installation Tiers

Tier	Skills	Agents	Commands	Hooks	Use Case
complete	91	20	12	93	Full AI-assisted development (default)
standard	91	0	12	93	All skills, spawn agents manually
lite	10	0	5	93	Essential skills, minimal context
hooks-only	0	0	0	93	Safety guardrails only

After installation, skills load automatically based on task context.

---

Skill Categories

AI & LLM (ai-llm/ - 19 skills)

Focus: Building AI-powered applications with RAG pipelines, embeddings, multi-agent orchestration, and cost optimization through caching.

Skill	Description
agent-loops	Agentic workflows with ReAct, tool use, and autonomous loops
rag-retrieval	RAG pipelines, chunking strategies, retrieval patterns
embeddings	Embedding models, vector operations, similarity search
function-calling	LLM function/tool calling patterns for OpenAI, Anthropic, Ollama
ollama-local	Local LLM inference with Ollama, model selection, optimization
multi-agent-orchestration	Coordinating multiple AI agents for complex tasks
prompt-caching	Anthropic/OpenAI prompt caching for cost reduction
semantic-caching	Semantic similarity caching with Redis/vector stores
cache-cost-tracking	LLM cost tracking and optimization
llm-streaming	Streaming responses, SSE, token-by-token output
llm-evaluation	Evaluation frameworks, benchmarks, quality metrics
llm-testing	Testing LLM applications, mocking, deterministic tests
llm-safety-patterns	LLM security, prompt injection prevention, context separation
langfuse-observability	LLM tracing, evaluation, prompt management, and cost tracking
hyde-retrieval	HyDE (Hypothetical Document Embeddings) for vocabulary mismatch
query-decomposition	Multi-c