Skip to content

skill.md

Latest commit

 

History

History
76 lines (58 loc) · 2.75 KB

skill.md

File metadata and controls

76 lines (58 loc) · 2.75 KB
Raw

Using expert-mentor as an agent skill

The repository root is a valid skill: SKILL.md tells an agent when to use expert-mentor and how to run it. That makes the whole thing usable inside Claude Code and opencode, not just as a CLI.

Install

mentor skill

Works whether mentor came from a git checkout or from pip install/pipx install. From a checkout it symlinks the whole repo into the standard locations (so it stays live as you edit the repo):

~/.claude/skills/expert-mentor       -> <repo>
~/.config/opencode/skills/expert-mentor -> <repo>

From a pip/pipx install (no repo to point at) it copies the bundled SKILL.md, templates/, and references/ into the same locations instead. Add --dir PATH to install elsewhere, or --force to replace a directory that already exists there for another reason. ./install.sh still does the same thing as part of its full one-shot setup (launcher + PATH + skill).

Restart the host (opencode loads skills at startup) so it picks up the new skill.

When it activates

The skill's description front-loads the trigger keywords, so an agent loads it when the user wants to learn, be taught, or be mentored in a field — or wants to turn a provider or local LLM into a teacher. Examples:

  • "teach me quantum computing"
  • "be my Rust mentor"
  • "explain macroeconomics like a professor"
  • "make Claude act as an expert tutor for contract law"

What the agent does

  1. Collects the essentials — field (required), plus level, goal, context, language, provider (inferred if absent).
  2. Chooses a delivery mode — teach now inline, produce a reusable prompt, configure a provider/local model, or design a curriculum.
  3. Generates with the CLI — e.g. mentor --field "Rust" --provider claude --emit json.
  4. Runs the session following the teaching contract, or hands the prompt off.

Alternative: project or global registration

Instead of symlinks you can register the skill path in your opencode config:

{
  "$schema": "https://opencode.ai/config.json",
  "skills": {
    "paths": ["~/projects"]
  }
}

The loader scans for **/SKILL.md inside configured paths, so ~/projects/expert-mentor/SKILL.md is found as the expert-mentor skill.

Files the agent reads

File Role
SKILL.md trigger + workflow
templates/mentor_system_prompt.md the master prompt
templates/compact_prompt.md small-model variant
references/pedagogy.md teaching methods
references/fields.md field profiles
references/providers.md provider recipes