Skip to content

Skills

A skill is a folder with a SKILL.md file — instructions the model can pull in when a task calls for them — plus any reference files it needs. It’s the same format Claude uses, so skills you already have (for example in ~/.claude/skills) work in Moka with any model.

  • Directorybrand-voice/
    • SKILL.md required: frontmatter + instructions
    • examples.md optional reference files
    • Directorytemplates/
      • reply.md
SKILL.md
---
name: brand-voice
description: House style for customer-facing replies. Use whenever writing a message to a customer.
---
# Brand voice
- Warm, plain English. No jargon.
- Structure: acknowledge → explain in one sentence → concrete next step.
- Sign off with "— The Moka team ☕".
See `examples.md` for good and bad replies.

Skills use progressive disclosure, so dozens of them cost almost no context:

  1. The system prompt lists each enabled skill’s name and description only.
  2. When a request matches, the model calls the built-in load_skill tool and receives the full SKILL.md body plus a list of the skill’s files.
  3. If the instructions reference other files, the model calls read_skill_file to read them.

Both calls show up in the chat and the inspector, so you can see exactly when and why a skill was used.

Settings → Skills → Add skill:

  • Folder — a path to a skill folder (or directly to a SKILL.md). Absolute, ~/…, or relative to your config file. Click Preview to see exactly what the model will get.
  • Scan — point at a folder of skills (e.g. ~/.claude/skills) and add them all at once.
  • Write — author a skill inline; it’s stored in moka.json.

Enable or disable skills per workspace with the switch at the top of each skill, or in Workspaces.

  • Files are read fresh on every use — edit SKILL.md and ask again, no restart.
  • read_skill_file can only read inside the skill’s folder; path traversal is rejected.
  • Skills that include scripts: Moka exposes the files to the model to read, but doesn’t execute them. Pair a skill with an MCP server (for example a shell or code-execution server) if you need execution.