Skills

The problem with typing a new hyper-detailed prompt is the fundamental problem all AI aims to solve: we're lazy af and don't like to type.

The problem with using AGENTS.md and other documentation for all your reusable prompts is that there are a lot of prompts you'll want to reuse often, but not in literally every session.

Skills are prompts that you reuse frequently, but not in every session.

A skill is a directory with a Markdown file and optional additional assets, like scripts or other resources. The agent can load the skill automatically, or you can load it manually. For example, I have a talk-to-database skill. It tells the agent how to connect to my production database safely (and in read-only mode) so that it can answer questions about the data without having to re-explain all the steps specific to my setup each time.

Add type checking to Sloperator's validation skill so the agent checks for type errors whenever you use it.

1. Tell OpenCode to read .agents/skills/validate/SKILL.md.
2. Tell it to add uv run ty check as the fourth check, after the tests, formatting check, and linter. Have it update the skill's description to mention type checking too.
3. Invoke @validate. Watch it run all four checks, then summarize the results.

Tell OpenCode to run and submit the CLI tests from the project root.