Claude Trading Bot: What Claude Can (and Can't) Do, Plus a Safe Starter

You're helping me build my first trading bot. Safety beats speed.

Step 0 (do this first, then stop and show me):
- Create CLAUDE.md containing the BotCookery Safety Pack rules I paste below.
- Create .gitignore covering .env, .env.*, *.pem, *.key, secrets/, KILL,
  logs/, .venv/ (but keep .env.example tracked).
- Create .claude/settings.json with permissions.deny for
  Read(**/.env), Read(**/.env.*), Read(**/*.pem), Read(**/*.key),
  Read(./secrets/**) and Bash(git push *). (If you are not allowed to write
  under .claude/, tell me and I will approve it; do not work around it.)
- Create limits.json with: max_contracts_per_order 1,
  max_open_exposure_usd 5, max_daily_loss_usd 2, max_orders_per_minute 3.
- Create .env.example with blank KALSHI_API_KEY_ID and KALSHI_PRIVATE_KEY_PATH.

Step 1: Build a Kalshi DEMO bot in Python 3.11+ (requests, cryptography,
pytest only) using only endpoints from https://docs.kalshi.com (fetch and
read the pages before coding):
- Base URL https://external-api.demo.kalshi.co/trade-api/v2 (hardcode demo;
  production must be a separate, deliberate change later).
- Signed requests: headers KALSHI-ACCESS-KEY, KALSHI-ACCESS-TIMESTAMP (ms),
  KALSHI-ACCESS-SIGNATURE = base64 signature of
  timestamp + METHOD + FULL path from the API root, e.g.
  "/trade-api/v2/portfolio/events/orders", with no query string. Support
  both key types: RSA (RSA-PSS, SHA-256, salt length = digest length) and
  Ed25519 (sign the message directly).
  Key id from KALSHI_API_KEY_ID; key file path from KALSHI_PRIVATE_KEY_PATH.
  Never read the key file contents into logs or output.
- If no credentials are set, run in DRY-RUN: read public market data, log what
  would be sent, send nothing.
- Watch one ticker from config.toml. Read the best YES ask from
  GET /markets/{ticker} (yes_ask_dollars, yes_ask_size_fp; fixed-point dollar
  strings). Rule: if the best YES ask is at or below my configured max price
  and size >= 1, propose a 1-contract limit bid at that price. (To find a
  ticker with a live ask on demo:
  GET /markets?status=open&mve_filter=exclude&limit=1000 and look for
  yes_ask_dollars between 0 and 1.)
- risk.py checks KILL file (or BOT_KILL=1) + limits before every order:
  max contracts, open exposure (sum of price*count of orders sent), daily
  loss (placeholder input defaulting to 0 until fills are wired), orders per
  minute.
- Orders via POST /portfolio/events/orders with body:
  ticker, side "bid" (bid = buy YES), count as "1.00", price as "0.3500",
  client_order_id (uuid4), time_in_force good_till_canceled,
  self_trade_prevention_type taker_at_cross.
- cancel_all.py -> DELETE /portfolio/events/orders (exit with a clear
  message if no credentials).
- Log every decision to logs/decisions.jsonl.
- Tests for risk.py (every rule), the signing message (query string stripped,
  full path), and a signature verified with the public key; use a fake HTTP
  session so tests need no network or keys.

Explain each file in plain English when you're done, and list anything
you weren't sure about.

[PASTE THE SAFETY PACK RULES FILE HERE]