# Mylestech OpenCode — Agent Instructions

You are coding on the Mylestech internal platform. All model calls route through
the LiteLLM gateway (`auto` picks the best model your role allows). A central
**skills catalog** is installed at `~/.config/opencode/skills`.

## Using skills

- Before starting any non-trivial task, scan the skills catalog for a matching
  skill. `WORKING_SET.md` lists the curated coding/efficiency skills; the repo
  root holds the full catalog (472 skills).
- A skill lives in `<skill-name>/SKILL.md` with YAML front-matter (`name`,
  `description`) and optional `references/`. Read the `SKILL.md` when its
  description matches your task, then follow it.
- High-value defaults: `ask-questions-if-underspecified` (clarify first),
  `brainstorming` (design before building), `verification-before-completion`
  (prove it works), `find-skills` (locate the right skill).

## Keeping skills current

The catalog is a git clone of the platform's `skills` repo. To update:

    git -C ~/.config/opencode/skills pull

## Platform norms

- Prefer local models for routine work; the gateway escalates to cloud/premium
  only when your role/grant allows. Do not hardcode model names — use `auto`.
- Never commit secrets or credential files. Keep your `.env` private.