Guide

Skills

Larik supports Agent Skills: folders containing a SKILL.md with YAML frontmatter, plus optional scripts and resources.

---
name: release-notes
description: Write release notes from git history. Use when asked for a changelog or release notes.
---

Collect commits since the last tag with `git log $(git describe --tags --abbrev=0)..HEAD` ...

• Discovery, lowest to highest precedence:

  1. ~/.agents/skills, ~/.claude/skills, ~/.config/larik/skills
  2. .agents/skills, .claude/skills, .larik/skills in each directory from the repo root down to the working directory

  Existing Claude Code skills work as-is, symlinked skill folders are followed, and a higher-precedence skill with the same name overrides the lower one (/skills shows what was shadowed).

• Progressive disclosure: only each skill's name: description is in the system prompt. The model loads the full instructions with the read-only skill tool when a task matches, and reads bundled files with the normal tools. The index is rebuilt at each fresh context (a new session or /clear) and stays fixed within it, so prompt caches stay valid. Add or edit a skill, then /clear to use it.

• Running a skill yourself: type /<skill-name> [args], in the TUI or with -p. $ARGUMENTS in the body is replaced with the args, and $1 to $9 with the args one by one (split on spaces). Otherwise the args are appended.

• Frontmatter flags: disable-model-invocation: true keeps a skill out of the model's index, so it only runs when you invoke it. user-invocable: false hides it from /.

• Skills are instructions, like AGENTS.md. Anything a skill asks the model to run still goes through the normal permission checks.

#Custom commands

Claude Code–style custom commands work too: a Markdown file whose name is the command.

---
description: Fix a GitHub issue
argument-hint: [issue-number] [priority]
---

Fix issue #$1 with priority $2. Current branch: !`git branch --show-current`

• Where: ~/.claude/commands and ~/.config/larik/commands for yours; .claude/commands and .larik/commands from the repo root down to the working directory for the project's. A file in a subdirectory still takes its own name (frontend/review.md is /review).
• Running one: type /<name> [args]. The palette lists commands with their argument-hint, and /skills lists them with the skills. Arguments work as for skills.
• Frontmatter is optional. Without a description, the first line of the file describes the command in the palette. Only a command with a description goes in the model's index, where it can load it with the skill tool; disable-model-invocation: true keeps a described one out.
• Skills win: a skill and a command with the same name resolve to the skill, and built-in commands such as /model always win.
• !`command` runs the command when you invoke the command file and puts its output in its place, in the sandbox when there is one. It runs only when your permission rules would allow it without asking (sandboxed commands, the safe list, allow rules); otherwise it is left out with a note, since a command file can come from a repository. Claude Code's allowed-tools frontmatter is ignored for the same reason. model is ignored too.
• New or edited commands appear after /clear or in a new session, like skills.