Develop Agent

When to Use

• User says "create an agent that..."
• User runs /develop-agent "description"
• User describes wanting a specialized agent for a task
• User wants to formalize a workflow into an agent

Workflow

Step 1: Initial Interview

Ask these agent-specific questions to clarify scope:

Scope & Purpose:

• What is the primary function of this agent?
• What problem does it solve?
• Who will use it (developer, analyst, end user)?

Input & Output:

• What inputs will the agent receive?
• What outputs should it produce?
• What format should outputs be in?

Tools & Capabilities:

• What tools does it need? (Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Skill)
• Should it spawn other agents?
• Any tools it should NOT have?

Constraints:

• What should it NOT do?
• Any scope boundaries?
• What decisions should it make vs. ask the user?

Narrow Scope Validation:

Apply the Three Tests for Narrow Scope:

1. Trigger Test: Does this agent have exactly one trigger condition?

   • If writing "when X or when Y," you have two agents in one
   • Each agent should have ONE specific situation

2. Action Test: Does this agent produce exactly one type of output?

   • If actions can fire independently, they should be separate agents
   • One output type per agent

3. Failure Test: If one part fails, does the rest still make sense?

   • Independent failure modes indicate separate agents
   • Contained failure scope

Complexity Detection:

If answers reveal complex functionality (multiple workflows, intricate logic, many edge cases):

• Spawn functional-analyst for deep requirements analysis
• Wait for functional-analyst to complete
• Continue with its analysis document

If straightforward (clear, focused task):

• Continue with the interview and create analysis directly

Step 2: Create Analysis Document

ALWAYS create an analysis document in the idea folder:

ideas/{agent-name}/analysis/functional.md

Analysis document structure:

# {Agent Name} - Functional Analysis

**Date:** YYYY-MM-DD
**Analyst:** [skill or functional-analyst]

## Purpose

[One paragraph describing what the agent does and why]

## Scope

### In Scope
- [Task 1]
- [Task 2]

### Out of Scope
- [Excluded task 1]
- [Excluded task 2]

## Inputs

| Input | Type | Description |
|-------|------|-------------|
| [input1] | [type] | [description] |

## Outputs

| Output | Type | Location |
|--------|------|----------|
| [output1] | [type] | [path] |

## Tools Required

| Tool | Usage | Risk Level |
|------|-------|------------|
| [tool1] | [what it's used for] | [read/modify/delete] |

## Constraints

- [Constraint 1]
- [Constraint 2]

## Workflow

1. [Step 1]
2. [Step 2]
3. [Step 3]

## Example Scenarios

### Scenario 1: [Name]
**Input:** ...
**Expected Output:** ...

## Decisions Made

| Decision | Rationale |
|----------|-----------|
| [decision] | [why] |

## Related Agents

- [agent1] - [relationship]
- [agent2] - [relationship]

## Scope Validation

- Trigger Test: [single trigger condition]
- Action Test: [single output type]
- Failure Test: [contained failure scope]

Step 3: Design Agent Structure

Color Selection

Step 4: Create Agent Definition

Create the agent file:

ideas/{agent-name}/artifacts/agent/{agent-name}.md

Five-Section System Prompt Framework

See references/system-prompt.md for the complete framework. Every agent must include:

1. Identity and Role
2. Capabilities and Constraints
3. Tool Instructions
4. Output Format
5. Guardrails and Error Handling

Agent Frontmatter Template

---