# 1. Run the installer (no global install)
npx @ctxr/kit@latest

# 2. Pick a memory backend (one prompt, one decision)
#    Wiki  → https://github.com/ctxr-dev/llm-wiki-memory#install
#    RAG   → https://github.com/ctxr-dev/memory#install

# 3. Add the skill almost everyone wants
npx @ctxr/kit@latest install @ctxr/skill-code-review

That is the whole start. Everything below is opt-in.

Self-improving memory: your agent recalls past lessons before it works and saves a new one the moment you correct it, so the same mistake does not recur. Both options expose the same tools and behavior, so you can switch later. Install one, not both.

	📒 Wiki recommended	🧬 RAG
Repo
Stores as	git-versioned markdown in your repo	local Dify vector store
Infra	none (Node + git)	Docker + Dify
Best for	solo, small, and medium projects; offline; low overhead	large corpora; teams; retrieval precision at scale
One-prompt AI install	llm-wiki-memory#install	memory#install

Default to Wiki. Choose RAG when the corpus is large, several people or agents share one store, or Docker is already in your stack. Not ready for persistent memory? Skip it. Nothing else depends on it.

%%{init: {'theme':'base','themeVariables':{'background':'#0B1F23','primaryColor':'#102E32','primaryTextColor':'#F0E8D5','primaryBorderColor':'#E5C547','secondaryColor':'#163A3D','tertiaryColor':'#0A1E22','lineColor':'#3DBE9C','clusterBkg':'#0A1E22','clusterBorder':'#1A3A3A','fontFamily':'ui-monospace, JetBrains Mono, monospace'}}}%%
flowchart TB
    user(["👩‍💻 You"])
    agent{{"🤖 Your agent\n Claude Code · Codex · Cursor · any MCP client"}}

    subgraph installer ["📦 Installer"]
      kit["@ctxr/kit"]
    end

    subgraph skills ["🛠️ Skills"]
      sCR["skill-code-review"]
      sFE["skill-frontend-excellence"]
      sLW["skill-llm-wiki"]
    end

    subgraph agents ["🧑‍🚀 Agents"]
      aSE["agent-staff-engineer"]
      aCE["agent-codebase-explorer"]
      aPR["agent-plan-reviewer"]
      aIA["agent-implementation-auditor"]
    end

    subgraph memory ["🧠 Memory"]
      mW["llm-wiki-memory"]
      mR["memory (RAG)"]
    end

    subgraph integ ["🔌 Integrations"]
      mcp["@ctxr/mcp-github"]
      meth["github-dev-methodology"]
    end

    user --> agent
    agent --> kit
    kit -.installs.-> skills
    kit -.installs.-> agents
    agent --> memory
    agent --> integ
    aSE -.uses.-> sLW
    aSE -.reviews via.-> sCR

Package	Use it to	Badges
@ctxr/kit	install, update, and scaffold skills and agents (npx @ctxr/kit@latest, no global)

Install any skill with npx @ctxr/kit@latest install @ctxr/<name>.

Skill	Use it to	Badges
@ctxr/skill-code-review	multi-specialist review with a GO / NO-GO verdict
@ctxr/skill-frontend-excellence	ship fast, accessible, distinctive web UI (Lighthouse 95+ mobile, 99+ desktop)
@ctxr/skill-llm-wiki	make your agent read docs and code token-efficiently

Install with npx @ctxr/kit@latest install @ctxr/<name>.

Agent	Use it to	Badges
@ctxr/agent-staff-engineer	drive a ticket to an open PR, then hand off before merge (pulls in skill-llm-wiki, reviews via skill-code-review)
@ctxr/agent-codebase-explorer	read-only "where is X / what references Y" search subagent, capped structured reports
@ctxr/agent-plan-reviewer	adversarially review a plan or design before you confirm it: gaps, blind spots, edge cases, infeasibilities
@ctxr/agent-implementation-auditor	post-build conformance audit: missed plan items, divergences, cross-implementation parity

💡 The codebase-explorer, plan-reviewer, and implementation-auditor are read-only by design: their tool surface is scoped to Read / Grep / Glob / Bash, so a flaky MCP connector cannot kill subagent init. Drop them in front of the staff-engineer loop or use them standalone.

Server	Use it to	Setup	Badges
@ctxr/mcp-github	structured GitHub tool calls instead of gh shell-outs	register it

Repo	Use it to	How	Badges
ctxr-dev/github-dev-methodology	a consistent GitHub issue and PR workflow, plus subagent orchestration	clone into your project; read it

Package	Use it to	Badges
@ctxr/fsm	author your own deterministic multi-agent workflow (usually transitive, via a skill or agent)

The ctxr stack plays nicely with other open skill and agent collections. A few we like:

The npx skills@latest add <owner>/<repo> pattern works for any GitHub repo that follows the skills layout. Use @ctxr/kit for ctxr packages, skills CLI for everything else, side by side.

All repos are MIT licensed and developed in the open at github.com/ctxr-dev.

• 🐞 Found a bug? Open an issue on the relevant repo.
• 💡 Have an idea? Discussions are on, fire away.
• 🛠️ Want to ship a skill or agent under @ctxr/? Start from @ctxr/kit and PR it.

_{Built in the open for agents that ship.}