Built-in Agents
Elftia provides an Agent gallery with pre-configured Agents for a variety of scenarios. You can use these Agents directly or create customized versions based on them.
Agent Gallery
The Agent gallery organizes preset Agents into categories by use case, making it easy to find the right one quickly.
Browsing Agents
- Click the Agent icon in the sidebar to open the Agent panel
- At the top you will see the Gallery / My Agents toggle tabs
- Select the "Gallery" tab to browse all preset Agents
- Use the search box or category filter to locate one quickly
Agent Categories
| Category | Description | Included Agent Types |
|---|---|---|
| Work | Work | Project management, task breakdown, meeting notes, weekly reports |
| Business | Business | Business analysis, market research, planning proposals |
| Tools | Tools | File processing, data conversion, automation scripts |
| Language | Language | Translation, proofreading, grammar checking, multilingual |
| Office | Office | Document writing, spreadsheet handling, presentations |
| Creative | Creative | Copywriting, story writing, brainstorming |
| Coding | Coding | Code generation, code review, bug fixing, testing |
| Education | Education | Knowledge Q&A, tutoring, learning plans |
Selecting and Using an Agent
- Click on an Agent card in the gallery
- View the Agent's details:
- Description — the Agent's capabilities and applicable scenarios
- Tool list — tools the Agent can use
- Permission mode — security level for tool execution
- Model — the LLM model the Agent uses
- Click "Use" to start a conversation
- You can also click "Favorite" to add the Agent to your quick-access list
Favorites and Ratings
- Favorite — click the star icon on an Agent card; favorited Agents appear at the top of the "My Agents" list
- Rate — after using an Agent, you can rate it to help other users discover quality Agents
Agent Configuration Format
Each Agent is essentially a Markdown file that uses YAML frontmatter to define configuration parameters, with the body serving as the system prompt.
File Structure
---
name: Code Review Assistant
description: Review code quality, identify potential issues, and suggest improvements
model: main
permissionMode: default
tools:
- Read
- Glob
- Grep
- ListDir
skills:
- code-standards
---
You are a professional code review assistant.
## Review Focus
1. Code quality and readability
2. Potential bugs and performance issues
3. Security vulnerabilities
4. Adherence to best practices
## Output Format
For each issue found, provide:
- Issue description
- Severity (High / Medium / Low)
- Suggested fix
Configuration Field Reference
| Field | Type | Required | Description |
|---|---|---|---|
name | string | Yes | Agent name |
description | string | Yes | Agent description (displayed on gallery cards) |
model | string | No | Model selection (supports aliases, default: main) |
permissionMode | string | No | Permission mode (default: default) |
tools | string[] | No | Allowed tool list (empty = inherit all) |
skills | string[] | No | Names of skills to auto-load |
Model Aliases
The model field in Agent configuration supports the following aliases so configs are not tied to a specific model:
| Alias | Meaning | Description |
|---|---|---|
main | Main model | Uses the main model from the current session config (default) |
inherit | Inherited model | Same as main; inherits the parent Agent's model |
background | Background model | Uses the user's configured background model (lightweight) |
sonnet | Sonnet family | Switches the parent model to the Sonnet family (Anthropic only) |
opus | Opus family | Switches the parent model to the Opus family (Anthropic only) |
haiku | Haiku family | Switches the parent model to the Haiku family (Anthropic only) |
:::tip Use semantic aliases
It is recommended to use semantic aliases such as main, inherit, and background — they are not tied to any specific provider. sonnet, opus, and haiku are only effective when using Anthropic models.
:::
Permission Modes
| Mode | Description |
|---|---|
default | Sensitive tools (Bash, Write, Edit) require user confirmation |
acceptEdits | File edits are approved automatically; Shell and MCP tools still require confirmation |
bypassPermissions | All tools execute automatically without confirmation |
plan | Read-only mode; only read-only tools such as Read, ListDir, Glob, Grep are permitted |
Agent Storage Locations
Agent configuration files can be placed in the following locations:
| Location | Path | Scope |
|---|---|---|
| Project level | .claude/agents/*.md | Available only in the current project |
| Personal level | ~/.claude/agents/*.md | Available across all projects |
| Built-in | Bundled with the app | Available to all users |
Load priority: project level > personal level > built-in. If the same Agent name exists in multiple locations, the project-level version takes precedence.
FAQ
| Problem | Cause | Solution |
|---|---|---|
| An Agent is not visible in the gallery | Agent config file is not in the correct path | Check that the file is in the .claude/agents/ directory |
| Agent cannot use a specific tool | The tool is not listed in the tools field | Add the tool name to the frontmatter |
| Model alias not taking effect | Using an alias only supported by Anthropic | Switch to the main or background semantic alias |
| Custom Agent not visible in other projects | Saved to the project-level directory | Move it to ~/.claude/agents/ (personal level) |
Related Links
- Agent Overview — Agent system overview
- Creating Custom Agents — Build an Agent from scratch
- Tool Permissions & Security — Permission mode details
- Skill System — Configure skills for Agents