Skills

A skill is a folder with a SKILL.md in it: a name, a one-line description of when to use it, and instructions the agent follows when the task matches. It can carry scripts and reference files beside it. The format is an open standard (agentskills.io) that Claude Code, Codex, OpenCode, Gemini CLI, Cursor and others read, and OptiQ Code reads the same folders. A skill you already set up for one of them works here with no extra step.

Where OptiQ Code looks

In the project, then in your home directory:

.agents/skills/<name>/SKILL.md      # this project
.claude/skills/<name>/SKILL.md
~/.agents/skills/<name>/SKILL.md    # every project
~/.claude/skills/<name>/SKILL.md

When two folders hold a skill of the same name, the project's wins. /skills lists what was found, where each one came from, and anything skipped with the reason: a missing description, or a name that is not lowercase letters, digits and hyphens.

A skill

---
name: release-notes
description: Writes release notes from the git log. Use when asked for release notes or a changelog entry.
---

Read `git log` since the last tag. Write one line per user-visible change,
newest first, and leave out refactors and test-only commits.

Using one

Every skill's name and description goes into the system prompt. When a task matches, the agent calls the skill tool, which returns the full instructions and lists the skill's other files; it opens those with read_file when the instructions point at them. Only the descriptions cost context up front, and with many skills installed the list shortens itself to first sentences, then to names, so a local model's window is not spent on skills the task does not need.

To run a skill yourself, type its name as a command, in the terminal UI or in a headless prompt:

# in the terminal UI
/release-notes for v2.1

# headless
$ optiq code -p "/release-notes for v2.1"

A built-in command of the same name wins, so a skill called help does not replace /help.

Installing skills

Any installer that writes to these folders works. skills.sh installs from a GitHub repository, and anthropics/skills collects Anthropic's own:

$ npx skills add owner/repo

What a skill cannot do

Loading a skill only reads text. A script the skill asks for runs through bash and needs your approval like any other command. The standard's allowed-tools field is ignored, so a skill in a repository you cloned cannot approve its own commands.

The minimal preset reads no skills, so a benchmark run through it does not change with what you have installed. Plan mode reads them, since loading one changes nothing. To turn skills off, set "skills": false in ~/.optiq/code/config.json.