Agent Profile Creator

config.json

~/.aider-desk/agents/{name}/

{project}/.aider-desk/agents/{name}/

• Tool groups & schema:

  references/agent-profile-schema.md

• Tool approvals:

  references/tool-approval-guide.md

  +

  references/agent-profile-schema.md

• Subagent config:

  references/subagent-guide.md

• Examples:

  references/profile-examples.md

Simplified Q&A Process

Step 1: Understand Purpose

• Name (derived from purpose)
• Max iterations (default 250)
• Tool groups (based on purpose)
• Custom instructions (if applicable)
• Subagent config (always enabled, contextMemory: "off" by default)
• Tool approvals (based on needs)

Step 2: Provider/Model

• Split by first slash to get provider and model
• No validation - use as-is (correct obvious typos only)

Step 3: Optional Advanced Settings

• "What temperature should I use?" (0.1-1.0, optional)
• "Max tokens limit?" (optional)
• "Min time between tool calls in ms?" (optional)
• "Which MCP servers should be enabled?" (optional, default: none)

Step 4: Location

Step 5: Single Review & Confirm

• Name, provider/model
• Tool groups enabled
• Tool approvals summary (which are "never", which are "always")
• Subagent config
• Location

Step 6: Create

references/profile-examples.md

Tool Group Reference

• Power Tools: File system, shell, search (for agents that read/write files)
• Aider Tools: AI code generation via Aider (NOT default - see note below)
• Power Search: Advanced code search (for analysis/docs)
• Todo Tools: Task list management (for agents that track work)
• Memory Tools: Persistent information storage (for learning agents)
• Skills Tools: Access to project-specific skills (for agents using custom skills)
• Subagents: Task delegation to specialized subagents (for agents that can call others)

• If user wants coding: explain Aider (advanced refactoring) vs Power Tools (direct editing)
• Ask preference before enabling

Tool Approval Strategy

{group}---{tool}

• "never": Tools completely irrelevant (e.g.,

  power---bash

  for read-only agents)
• "always": Safe, essential tools (e.g., read operations for reviewers)

references/tool-approval-guide.md

Subagent Configuration

references/subagent-guide.md

• contextMemory

  : Default is

  "off"

  (fresh each time)
  • Use

    "full-context"

    only for specialized analysis agents (code review, security audit)
  • Use

    "last-message"

    for progressive builders (documentation, iterative tasks)

• systemPrompt

  : Specialized for agent's purpose

• invocationMode

  : automatic (specialized), on-demand (general)

• color

  : Relevant color (e.g., red=security, blue=power tools)

• description

  : Clear description for auto-invocation

File Structure

{profile-name}/
└── config.json

Validation

• Unique name
• Provider/model: Use as-is (correct obvious typos only)
• Tool keys:

  {group}---{tool}

  format
• Values: "always", "ask", or "never"
• Subagent enabled
• All tool keys from references

Resources

• references/agent-profile-schema.md

  - Complete schema

• references/subagent-guide.md

  - Subagent configuration guide (context memory modes)

• references/profile-examples.md

  - Examples

• references/tool-approval-guide.md

  - Tool approval config

• assets/templates/config.json.template

  - Template

• assets/examples/sample-profile.json

  - Example