SahilMund

Claude Agent Lab — AI skill for Claude Code

AI community

RAG-powered CLI code assistant with agentic task planning, MCP tool integration, semantic caching, and session memory — built incrementally as a system-design learning project, phase by phase with ful.

How to install Claude Agent Lab

This entry records only its repository, not the path inside it, so there is no exact command to give. Open SahilMund/claude-agent-lab and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Claude Agent Lab does

Alternatives in AI

Cursor Talk To Figma MCP — TalkToFigma: MCP integration between AI Agent (Cursor, Claude Code, Codex) and Figma, allowing Agentic AI to c 7k ★
Karpathy — An agentic Machine Learning Engineer 1.4k ★
Simulink Agentic Toolkit — The Simulink Agentic Toolkit gives your AI agent both the tools and the expertise to work effectively with Sim 987 ★

README

claude_agent_lab

A RAG-powered, agentic CLI code assistant — built phase by phase as a system-design learning project.

Each phase's design decisions, tradeoffs, and real bugs found along the way are documented as they happen, not silently. See `docs/prd.md` for the phase breakdown and `docs/progress.md` for what was built, fixed, and why.

Status

The core pipeline is built and verified against real infrastructure — indexing, retrieval, and the agent all tested against a live Qdrant instance and real MCP servers (not just "it imports"). The CLI (`main.py`) is the only interface.

What's here:

Config (config.py, config.yaml) — plain YAML, loaded once at import time. .env for secrets.
LLM/embeddings (llm/factory.py) — LangChain-based, provider chosen by config.yaml's llm.provider, no code change needed to switch: anthropic (default), openai, gemini, groq, or ollama (local, no API key). HuggingFaceEmbeddings for embeddings by default (runs locally, no API key needed).
Indexing & retrieval (context/indexers/, context/retrievers/) — tree-sitter-based code-aware chunking (15 languages), semantic (Chroma or Qdrant) or hybrid (Qdrant native sparse+dense) retrieval, chosen via config.yaml. Default: Qdrant + hybrid.
Agent (agent/) — LangChain's create_agent + LangGraph checkpointer-backed memory, with a search_codebase tool, filesystem tools (tools/filesystem_tools.py), a terminal tool (tools/terminal_tools.py), MCP tools (GitHub + filesystem servers), and skill-as-tool loading.
Memory (memory/) — session tracking (which conversation thread is "current") plus a SQLite-backed LangGraph checkpointer with automatic summarization once a conversation gets long. Plus long-term memory (memory/long_term.py) — cross-session facts/preferences in a separate Qdrant collection, retrieved and saved via recall/remember agent tools. See "Long-term memory" below.
MCP (mcp/) — connects to the serve