Skills

Skills are reusable automations invoked as slash commands. Each skill encapsulates a specialized prompt, allowed tools, and optionally lifecycle hooks. The agent executes the skill as if it were a complete instruction, maintaining session context.

How to invoke a skill

/skill-name [optional argument]

Examples:

/simplify
/loop 5m check failing deploys
/schedule
/update-config

Bundled skills (built-in)

The following skills are compiled into the Verboo Code binary and available in any installation.

`/update-config`

Modifies settings.json or settings.local.json safely. Includes a 7-step hook verification workflow that tests the hook before saving.

When to use: "add a permission for Bash(npx tsc *)", "configure a hook that formats files on save", "enable auto-compaction"

Available tools: only Read (doesn't edit files directly — uses the config tool)

`/simplify`

Analyzes recently changed code using 3 parallel agents, each with a different focus:

Agent 1 — Code Reuse: detects existing helpers being reinvented, logic duplication
Agent 2 — Quality: identifies redundant state, parameter sprawl, complex inline logic, unnecessary JSX, obvious comments
Agent 3 — Efficiency: flags unnecessary work, missed concurrency, expensive hot-path operations, memory leaks

After the 3 agents report, aggregates the findings and applies fixes.

When to use: after implementing a feature, before opening a PR

`/loop [interval] [prompt]`

Runs a prompt in a continuous loop. Two modes:

Fixed interval with prompt:

/loop 5m check failing CI builds
/loop 2h send progress report

Dynamic loop (no interval): The agent decides the next wake-up based on what it finds.

/loop babysit deploys until green

Maintenance loop (no argument): Works on whatever is pending — open PR, failing CI, unfinished work.

/loop

The minimum interval is 1 minute, maximum 1 hour. Accepted formats: 1m, 30s, 2h, 1 minute, 30 seconds.

`/schedule`

Creates, lists, and manages remote agents with cron expressions — agents that run autonomously on Verboo's infrastructure.

Flow:

Describe the agent's goal
The skill assembles the prompt
Define the cron (e.g., 0 9 * * 1-5 = weekdays at 9am)
Choose the model (default: claude-sonnet-4-6)
Confirm and create

/schedule

Requires Verboo account with remote agents permission.

`/keybindings-help`

Interactive guide to customize ~/.verboo/keybindings.json. Explains modifier syntax (ctrl, alt, shift, meta), special keys, chords (sequences), how to remove existing bindings, and validation rules.

`/batch`

Processes multiple tasks in batch via the Claude API (Batch API). Useful for running analyses on many files or generating content at scale.

Requires ANTHROPIC_API_KEY configured.

`/claude-api`

Helps develop applications using the Anthropic SDK. Guides on prompt caching, tool use, streaming, available models, and migration between versions.

External skills (`.verboo/skills/`)

You can create your own skills in Markdown and have them loaded automatically.

Directory structure

.verboo/
  skills/
    my-category/
      my-skill/
        SKILL.md          ← required
        reference.txt     ← optional context files

The SKILL.md filename defines the skill slug (parent directory is the name).

SKILL.md format

markdown

---
name: my-skill
description: "What this skill does — shown in /skills"
when-to-use: "Use when the user asks for X"
allowed-tools: [Bash, Read, Edit]
argument-hint: "[filename]"
model: claude-sonnet-4-6
user-invocable: true
---

You are an expert in [domain].

When invoked with `{{args}}`, do:
1. ...
2. ...

Additional context is in reference.txt.

Frontmatter fields

Field	Type	Description
`name`	string	Command slug (default: directory name)
`description`	string	Description shown in `/skills`
`when-to-use`	string	Hint for the model about when to invoke
`allowed-tools`	string[]	Tools the skill can use
`argument-hint`	string	Argument hint shown to the user
`model`	string	Specific model for this skill
`disable-model-invocation`	bool	If `true`, only injects the prompt without invoking the model
`user-invocable`	bool	If `false`, skill is hidden from the user
`hooks`	object	Skill-specific lifecycle hooks
`context`	`'fork'`	Runs in isolated context (fork)
`paths`	string	Gitignore-style patterns for conditional activation
`effort`	s/m/l/xl	Expected effort level

Conditional path-based activation

Skills with paths: in their frontmatter only become active when matching files are touched in the session:

markdown

---
name: review-tests
description: "Jest test review"
paths: "**/*.test.ts\n**/__tests__/**"
---

This avoids cluttering /skills with skills irrelevant to the current context.

Loading precedence

When skills have the same name, the following precedence order applies (higher wins):

Managed — skills managed by the Verboo platform
User — ~/.verboo/skills/
Project — .verboo/skills/ in the project
Legacy — .verboo/commands/ (deprecated)

Skill locations

Verboo Code looks for skills in:

~/.verboo/skills/ — global user skills
.verboo/skills/ — project skills (walk-up to $HOME)
Directories added with --add-dir

Skills via MCP

MCP servers can expose skills that are loaded automatically when the server is connected. Each MCP tool can be invoked as a skill if the server exports a skill prefix in its metadata.

See MCP for more details.