Agents

Agents are specialized sub-agents that handle specific tasks autonomously.

Canonical Format

type Agent struct {
    Name         string   `json:"name"`
    Description  string   `json:"description,omitempty"`
    Instructions string   `json:"instructions,omitempty"`
    Model        string   `json:"model,omitempty"`
    Tools        []string `json:"tools,omitempty"`
    Skills       []string `json:"skills,omitempty"`
}

Creating an Agent

Using Go

package main

import (
    "github.com/agentplexus/assistantkit/agents/core"
)

func main() {
    agent := core.Agent{
        Name:        "security-scanner",
        Description: "Scans code for security vulnerabilities",
        Instructions: `You are a security expert specializing in code review.

Your task is to analyze code for security vulnerabilities:

1. Check for OWASP Top 10 vulnerabilities
2. Look for hardcoded secrets
3. Identify insecure configurations
4. Find authentication/authorization issues

Always explain the risk and provide remediation steps.`,
        Model: "sonnet",
        Tools: []string{"Read", "Grep", "Glob"},
    }
}

Using YAML

Create agents/security-scanner.yaml:

name: security-scanner
description: Scans code for security vulnerabilities
instructions: |
  You are a security expert specializing in code review.

  Your task is to analyze code for security vulnerabilities:

  1. Check for OWASP Top 10 vulnerabilities
  2. Look for hardcoded secrets
  3. Identify insecure configurations
  4. Find authentication/authorization issues

  Always explain the risk and provide remediation steps.
model: sonnet
tools:
  - Read
  - Grep
  - Glob

Agent Options

Field	Description	Required
name	Agent identifier	Yes
description	Short description	Yes
instructions	Detailed behavior instructions	Yes
model	Model to use (sonnet, opus, haiku)	No
tools	Available tools	No
skills	Skills the agent can use	No

Assistant Support

Assistant	Agents Support
Claude Code	Yes
Gemini CLI	No
OpenAI Codex	No
AWS Kiro	Yes

Assistant-Specific Output

Claude Code

Agents are JSON files in agents/:

{
  "name": "security-scanner",
  "description": "Scans code for security vulnerabilities",
  "instructions": "You are a security expert...",
  "model": "sonnet",
  "tools": ["Read", "Grep", "Glob"]
}

AWS Kiro

Agents are stored in ~/.kiro/agents/:

{
  "name": "security-scanner",
  "description": "Scans code for security vulnerabilities",
  "prompt": "You are a security expert...",
  "model": "claude-sonnet-4",
  "allowedTools": ["read", "shell"]
}

Tool Mapping

Tools are mapped between canonical names and assistant-specific names:

Canonical	Claude Code	Kiro
Read	Read	read
Write	Write	write
Edit	Edit	edit
Bash	Bash	shell
Glob	Glob	glob
Grep	Grep	grep

Model Mapping

Canonical	Claude Code	Kiro
sonnet	sonnet	claude-sonnet-4
opus	opus	claude-opus-4
haiku	haiku	claude-haiku-3.5

Examples

Documentation Agent

name: doc-generator
description: Generates comprehensive documentation
instructions: |
  You are a technical writer specializing in developer documentation.

  Your task is to generate documentation that is:
  - Clear and concise
  - Well-organized
  - Complete with examples
  - Appropriate for the target audience

  Use the project's existing documentation style if present.
model: sonnet
tools:
  - Read
  - Glob
  - Write

Test Generator Agent

name: test-generator
description: Generates unit tests for code
instructions: |
  You are a testing expert who writes comprehensive unit tests.

  For each function or method:
  1. Identify edge cases and boundary conditions
  2. Write tests for happy path scenarios
  3. Write tests for error conditions
  4. Ensure good code coverage

  Follow the project's existing test patterns and frameworks.
model: sonnet
tools:
  - Read
  - Write
  - Bash

Performance Analyzer Agent

name: perf-analyzer
description: Analyzes code for performance issues
instructions: |
  You are a performance optimization expert.

  Analyze code for:
  - Algorithmic complexity issues (O(n²), O(n!), etc.)
  - Memory leaks and excessive allocations
  - I/O bottlenecks
  - Concurrency issues
  - Database query optimization

  Provide specific recommendations with benchmarks when possible.
model: opus
tools:
  - Read
  - Grep
  - Bash

Invoking Agents

Claude Code

Agents are invoked via the Task tool:

Use the security-scanner agent to analyze the authentication module.

AWS Kiro

Agents are invoked with the /agent command:

/agent security-scanner analyze src/auth/