Agents & Skills

Build custom Agents for your specific use cases and leverage Skills to extend their capabilities.

Agents are the core actors in Vibe. They are autonomous entities capable of performing tasks, making decisions, and interacting with the environment. Agents can be customized for specific use cases and extended with Skills to enhance their capabilities.

Agents can be selected using the --agent flag when starting vibe:

vibe --agent plan
vibe --agent auto-approve

vibe --agent plan
vibe --agent auto-approve

Or with Shift+Tab in the Interactive mode.

We provide a set of built-in agents you can leverage out of the box:

• Agents:
  • default: Requires approval for tool executions.
  • plan: Read-only agent for exploration and planning (auto-approve).
  • accept-edits: Auto-approves file edits only.
  • auto-approve: Auto-approves all tool executions.
• Subagents:
  • explore: Read-only subagent for codebase exploration.

Create custom agent profiles in ~/.vibe/agents/ by creating a .toml file, learn more about Vibe configuration here:

# ~/.vibe/agents/custom.toml
display_name = "Custom Agent"
description = "My custom agent description."
safety = "neutral"  # Options: "safe" | "neutral" | "destructive" | "yolo"
auto_approve = true
enabled_tools = ["read_file", "grep"]

# ~/.vibe/agents/custom.toml
display_name = "Custom Agent"
description = "My custom agent description."
safety = "neutral"  # Options: "safe" | "neutral" | "destructive" | "yolo"
auto_approve = true
enabled_tools = ["read_file", "grep"]

The safety field changes the user input border color to indicate the agent's safety level and currently has 4 choices. Their only purpose is to visually inform the user of the safety level of the agent, but does not implement safety measures.
We recommend using this field together with appropriate tool permissions, such as Agents capable of editing/deleting files being classified as destructive, or Agents restricted to read-only tool calls being classified as safe.

Vibe supports subagents for delegating tasks. Subagents run independently and can perform specialized work without user interaction.
The task tool allows you to delegate work to subagents:

• The agent uses the task tool to delegate work to a subagent.
• Subagents run independently and return results as text to the agent.

Subagents are useful for:

• Parallel work: Run exploration or research tasks in the background
• Specialized tasks: Use different agent profiles for specific work
• Safety: Subagents run in-memory without saving session logs

Custom subagents can be created in ~/.vibe/agents/:

# ~/.vibe/agents/research.toml
display_name = "Research"
description = "Read-only subagent for research tasks"
safety = "safe"
agent_type = "subagent"
enabled_tools = ["grep", "read_file"]

# ~/.vibe/agents/research.toml
display_name = "Research"
description = "Read-only subagent for research tasks"
safety = "safe"
agent_type = "subagent"
enabled_tools = ["grep", "read_file"]

Agents can ask questions interactively using the ask_user_question tool:

• Supports multiple-choice or free-text input.
• Displays questions as tabs for multi-question scenarios.
• Automatically used when the agent needs clarification or validation.

Vibe's skills system allows you to extend functionality through reusable components. Skills can add new tools, slash commands, and specialized behaviors.

Vibe follows the Agent Skills specification for skill format and structure.

Skills are defined in directories with a SKILL.md file containing metadata in YAML frontmatter. For example, ~/.vibe/skills/code-review/SKILL.md:

---
name: code-review
description: Perform automated code reviews
license: MIT
compatibility: Python 3.12+
user-invocable: true
allowed-tools:
  - read_file
  - grep
  - ask_user_question
---

# Code Review Skill

This skill helps analyze code quality and suggest improvements.

---
name: code-review
description: Perform automated code reviews
license: MIT
compatibility: Python 3.12+
user-invocable: true
allowed-tools:
  - read_file
  - grep
  - ask_user_question
---

# Code Review Skill

This skill helps analyze code quality and suggest improvements.

Vibe discovers skills from multiple locations:

1. Global skills directory: ~/.vibe/skills/
2. Local project skills: .vibe/skills/ in your project
3. Custom paths: Configured in config.toml

skill_paths = ["/path/to/custom/skills"]

skill_paths = ["/path/to/custom/skills"]

Enable or disable skills using patterns in your configuration:

# Enable specific skills
enabled_skills = ["code-review", "test-*"]

# Disable specific skills
disabled_skills = ["experimental-*"]

# Enable specific skills
enabled_skills = ["code-review", "test-*"]

# Disable specific skills
disabled_skills = ["experimental-*"]

Skills support the same pattern matching as tools (exact names, glob patterns, and regex when using the re: prefix).