# Instructions for AI coding agents — rinatsultan.com

This repository deploys the personal site of Rinat Sultanov on a shared VPS. Its live MCP/A2A service is **not** the AI Gateway.

## Boundaries

- Do not modify `ai-gateway-site/` when working on `rinatsultan.com`, unless the user explicitly requests that project.
- Do not activate DeepSeek, create paid model calls, publish client-specific generated pages, send prospect messages, create orders, process payments or enable payment protocols without separate explicit authorization and configured credentials.
- No production equipment actions, 1C connections or hidden customer data may be exposed through public protocol endpoints.
- Do not claim compliance with a protocol based only on a JSON file. Implement real behavior and tests. If a feature is not ready, advertise it as disabled or experimental.
- Do not commit secrets, access tokens, credential-bearing `.env` files, private user data, or real customer orders.
- The VPS hosts other production services. Do not change `/usr/bin/node`, global npm, nginx configuration for other virtual hosts, other systemd units, or firewall.
- Keep MCP and A2A backwards-compatible. New public actions must default to read-only, validate inputs, limit resources and return explicit errors.
- Never publish unverified company metrics, certificates, customers or financial ROI. The measured sales-agent case has explicit caveats.

## Development

- Node: isolated 24 LTS, service in `rinatsultan-mcp/`; Node deps pinned in `package-lock.json`.
- Run `npm ci --ignore-scripts`; `npm run test:production`.
- Run `python3 scripts/seo_validate.py` and `python3 -m unittest discover -s scripts -p 'test_*.py'`.
- Check Nginx with `nginx -t` on the VPS before reload.
- Publish via GitHub Actions workflow `deploy.yml` only after tests, inspect full workflow status and production smoke tests.
- A failed deploy may still have modified the VPS; verify and recover rather than reporting a clean rollback without evidence.
- `README.md`, `docs/monitoring.md` and `agent-lab/index.md` should reflect real deployed capabilities.

## Skills

See `.agents/skills/manufacturing-ai-diagnostic/SKILL.md` and
`.agents/skills/industrial-ai-pilot/SKILL.md` for reusable, independently reviewable methodologies.