Create Agents

Create role-specific Claude Code agent .md files tailored to the detected tech stack. Write files to .claude/agents/<name>.md.

Agent Roles Table

Role	Tools	Match When
frontend-dev	Read, Write, Edit, Bash, Glob, Grep	UI framework deps OR `.svelte`/`.vue`/`.jsx`/`.tsx` files
backend-dev	Read, Write, Edit, Bash, Glob, Grep	Server framework deps (express, fastapi, django, flask, gin, rails, laravel, phoenix, hono, etc.)
aws-architect	Read, Write, Edit, Bash, Glob, Grep	`cdk.json` OR `aws-cdk-lib` OR CloudFormation/SAM OR significant `@aws-sdk/*`
ai-engineer	Read, Write, Edit, Bash, Glob, Grep	`.claude-plugin/` OR SKILL.md authoring OR AI/LLM deps
mcp-developer	Read, Write, Edit, Bash, Glob, Grep	`@modelcontextprotocol/sdk` in deps OR MCP server code
devops	Read, Write, Edit, Bash, Glob, Grep	CI/CD configs OR Dockerfiles OR deployment scripts
dba	Read, Write, Edit, Bash, Glob, Grep	Database schemas, migration files, ORM deps
terraform-engineer	Read, Write, Edit, Bash, Glob, Grep	`*.tf` OR `terraform/` OR `pulumi/`
test-automator	Read, Write, Edit, Bash, Glob, Grep	Test framework deps (vitest, jest, playwright, pytest, etc.)
python-dev	Read, Write, Edit, Bash, Glob, Grep	Python primary (`pyproject.toml`/`requirements.txt` + `*.py`)
rust-dev	Read, Write, Edit, Bash, Glob, Grep	Rust primary (`Cargo.toml` + `*.rs`)
go-dev	Read, Write, Edit, Bash, Glob, Grep	Go primary (`go.mod` + `*.go`)
code-reviewer	Read, Grep, Glob, Bash	Always
security-auditor	Read, Grep, Glob, Bash	Always

From Preferred Installables

Preferred agents arrive with name and content (full markdown with frontmatter). Write directly to .claude/agents/<name>.md. Skip prompt generation and role matching.

Selection Rules

•Always include code-reviewer and security-auditor
•Always include at least one implementation agent matching the project
•Cap at 6 agents total
•Prefer more specific roles over generic ones (e.g., frontend-dev over python-dev if the project is a Python web app with a frontend)

Agent Prompt Format

Use frontmatter with these fields:

markdown

---
name: <role>
description: <one-line description tailored to project stack>
tools: <comma-separated tool list from table>
model: sonnet
memory: <project for implementation agents, user for review-only agents>
---
<role-specific prompt tailored to detected tech stack>

Optional frontmatter fields: skills (list of installed skill names), mcpServers (list of MCP server names).

Prompt Guidelines

•Adapt prompts to the detected tech stack (e.g., a frontend-dev for Svelte gets Svelte 5 runes guidance, not React hooks)
•Reference SPEC.md/PLAN.md if they exist in the project
•Include project-specific context: file paths, conventions, key dependencies
•Use sonnet as default model
•Use project memory for implementation agents, user for review-only agents (code-reviewer, security-auditor)
•Keep prompts concise — every instruction should earn its place

Example

markdown

---
name: frontend-dev
description: SvelteKit frontend specialist using Svelte 5 runes.
tools: Read, Write, Edit, Bash, Glob, Grep
model: sonnet
skills:
  - svelte5-development
memory: project
---
You are a senior SvelteKit developer specializing in Svelte 5 with runes ($state, $derived, $effect, $props).

Read SPEC.md for product requirements and PLAN.md for architecture decisions before starting any work.