agent-creator（Agent 创建者）

Agent Profile Creator

Create agent profiles stored as config.json in ~/.aider-desk/agents/{name}/ (global) or {project}/.aider-desk/agents/{name}/ (project-level).

Read all reference files before proposing a profile to ensure accuracy.

Q&A Process

Step 1: Understand Purpose

Ask user: "Describe the agent's purpose and what it should do."

Based on their response, internally propose all properties:

• 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

Ask: "Which provider/model? (format: provider/model, e.g., anthropic/claude-sonnet-4-5-20250929)"

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

Step 3: Optional Advanced Settings

ONLY ask if user specifically mentioned temperature, max tokens, rate limiting, or MCP servers:

• "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)

If not mentioned, skip this entire step.

Step 4: Location

Ask: "Global profile (all projects) or project-specific? (default: global)"

Step 5: Single Review & Confirm

Present one complete summary with all proposed properties. Ask:

"Here's your agent profile. Should I create it?"

Show:

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

DO NOT ask for confirmations on individual items. Only one final approval.

Step 6: Create

On user confirmation, generate the profile. Verify structure against references/profile-examples.md before creating files.

Tool Approval Strategy

Keys: {group}---{tool} (three dashes)

Default is "ask". Only set:

• "never": Tools completely irrelevant (e.g., power---bash for read-only agents)
• "always": Safe, essential tools (e.g., read operations for reviewers)

Only include tools that exist in references/tool-approval-guide.md.

Subagent Configuration

Every agent is a subagent (enabled: true). See references/subagent-guide.md for detailed guidance.

• 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

Minimal config.json Structure

{
  "name": "my-agent",
  "provider": "anthropic",
  "model": "claude-sonnet-4-5-20250929",
  "maxIterations": 250,
  "toolGroups": ["power", "todo"],
  "toolApprovals": {
    "power---bash": "ask",
    "power---read": "always"
  },
  "subagent": {
    "enabled": true,
    "contextMemory": "off"
  }
}

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