LLM Quick Start Skill

# Skill: Set up AgentRAM in this project

You are helping me wire AgentRAM (a persistent memory API for AI agents) into
this project. Work step by step. Only stop to ask me if something required is
missing or ambiguous. Otherwise proceed and report what you did.

## What AgentRAM is
A simple hosted key-value memory store: you save a value under a key and read it
back in any later session. Two integration paths:
- MCP server (preferred for Claude Code, Codex, Cursor): npx agentram-mcp
- SDK / REST (any Python or JS code): the agentram-sdk package

## Step 1 - Detect the environment
- Identify the OS, and whether this is a Claude Code / Codex / Cursor project or
  a plain Python / Node project.
- Report the versions of node (node -v) and python (python3 --version).
- The MCP path needs Node 18+. The SDK path needs Python 3.8+ or Node 18+.
  If a required runtime is missing or too old, tell me exactly what to install
  and stop.

## Step 2 - Get the API key
- Ask me for my AgentRAM API key if it is not already in the environment.
- It looks like agentram_... . Never print it back in full or commit it.
- Store it as the env var AGENTRAM_API_KEY (add to .env, and confirm .env is in
  .gitignore).

## Step 3 - Choose the path
- Claude Code / Codex / Cursor  -> set up the MCP server (Step 4a).
- Otherwise, or if I ask for code I can call directly -> set up the SDK (Step 4b).

## Step 4a - MCP server
- Add AgentRAM to this tool's MCP config (find the correct config file for THIS
  client; it differs between Claude Code, Codex, and Cursor):
      command: "npx"
      args: ["-y", "agentram-mcp"]
      env: { "AGENTRAM_API_KEY": "<read from environment, do not hardcode>" }
- Explain where you put it and how to reload the client so the tools appear.

## Step 4b - SDK
- Python: create a virtual environment first
  (python3 -m venv .venv && source .venv/bin/activate), then
  pip install agentram-sdk. Explain briefly why the venv matters (isolation).
- JS: npm install agentram-sdk.

## Step 5 - Smoke test (always do this)
- Write and run a tiny test that stores a value and reads it back, then prints a
  clear PASS or FAIL. Python example:
      import os
      from agentram import AgentRAM
      ram = AgentRAM(api_key=os.environ["AGENTRAM_API_KEY"], agent_id="setup-check")
      ram.store("smoke_test", "ok")
      assert ram.recall("smoke_test") == "ok"
      print("AgentRAM PASS")
- For the MCP path, instead confirm the client lists the AgentRAM tools, then
  have the agent store and recall one value.

## Common first-run issues (check these before asking me)
- 401 / auth error -> the key is wrong or not loaded from the env. Re-check that
  AGENTRAM_API_KEY is set in the current shell or session.
- npx cannot find agentram-mcp -> Node is older than 18, or you are offline.
  Check node -v.
- Python import error -> the venv is not activated, or you installed into the
  wrong interpreter. Confirm which python points into .venv.
- Recall returns nothing -> a store must succeed before a recall, and both calls
  must use the same agent_id.

When done, print a short summary: path chosen, where config and keys live, and
the smoke-test result.