Skip to content

Skills and commands

Skills and Commands share the composer’s / picker. A Skill supplies an instruction resource to a Session. A Command expands a saved prompt template into the message you send. Neither grants tools or changes the Session’s authority.

Open Settings → All projects → Skills. Choose Activation project to set modes for that project’s Sessions. All, Global, and This project filter where the Skill files come from; the activation rules you edit still apply only to the selected project.

Source Location
This project .agents/skills/<slug>/SKILL.md
Global ~/.agents/skills/<slug>/SKILL.md

The directory name is the Skill’s slug. If project and personal Skills share a slug, the project Skill wins. If a Command uses that name, it keeps /slug and the Skill uses /skill:slug.

Choose a Mode for each Skill:

Mode What it does
Auto Makes the Skill discoverable to the model and available from /. Its metadata can appear in the bounded Skills index.
Manual Keeps it out of the model’s index, but available when you invoke /slug.
Off Removes it from the model’s index and the / picker.

The Auto token estimate shows the metadata cost of discoverable Skills. The index has a size limit; it does not include full Skill bodies. A model reads SKILL.md when it activates a discoverable Skill. Invoking /slug explicitly includes the Skill’s instructions as a separate resource with your message.

Use a row’s Preview action to read its instructions without changing its mode. Reload reads the files again after you add or edit a Skill. You can select several rows to change their modes together.

Before you set a project mode, Volli follows the Skill’s frontmatter:

  • disable-model-invocation: true makes it manual-only.
  • user-invocable: false hides it from / while leaving model discovery enabled, unless model invocation is also disabled. Settings shows this as Model only (author).
  • Without either restriction, both model discovery and user invocation are enabled.

Choosing Auto, Manual, or Off overrides both author controls for that project. A warning beside a Skill identifies malformed or conflicting invocation declarations. Fix the file if its declaration does not match what you intended.

Select the project in Settings, then Commands. Configure → Commands opens the same page. Volli lists project and personal templates; a project Command takes precedence when both have the same name.

  1. Select New command.
  2. Enter Name and Prompt, and optionally Description.
  3. In Save to, choose This project or Personal.
  4. Select Create.
  5. In a chat composer, type / and choose the command.

Names accept letters, numbers, dashes, and underscores. A Command named review becomes /review.

Commands are Markdown files. The body replaces the slash command in your message. Use $1 for the first argument or $ARGUMENTS for all arguments. For example, this review template makes /review src/app.ts ask for a review of that file:

---
description: Review a file for bugs
---
Review $1 for correctness and missing tests.

Project Commands live in .volli/commands/<name>.md. Personal Commands live in Volli’s application-data commands directory and are available across projects. Write a markdown file provides the file-based creation path.

Distinguish Skills, Commands, and file references

Section titled “Distinguish Skills, Commands, and file references”

Use / for Skills and Commands. A Command expands into prompt text; a Skill arrives as a separate instruction resource alongside your message. Use @ for file references.