agent-factory_skill

Overview

This skill generates production-ready Claude Code agents and sub-agents with enhanced YAML frontmatter, tool access guidance, and MCP integration support. It provides consistent templates and validation to accelerate building, deploying, and discovering Claude agents across projects. The output is a single markdown agent file ready to install into .claude/agents or a personal agents directory.

How this skill works

Fill a simple prompt template with agent name, description, domain, tools, and MCP integrations, then run the template through Claude Code to produce a fully formed .md agent file. The generator injects rich YAML metadata (color, field, expertise, model, mcp_tools), enforces tool-access patterns, assigns safe execution patterns (parallel vs sequential), and validates the configuration against best practices. Generated agents reference recommended MCP servers when available and include checklists, invocation hints, and an output format to make them immediately usable.

When to use it

• Creating domain-specific agents (frontend, backend, data, security) for team workflows
• Standardizing agent metadata and discovery across multiple projects
• Prototyping multiple related agents quickly with consistent patterns
• Integrating agents with MCP servers like mcp__github or mcp__playwright
• Defining safe execution patterns to avoid parallelization issues

Best practices

• Keep each agent focused on a single responsibility to enable safe parallelization
• Use descriptive names and detailed descriptions so Claude can auto-invoke correctly
• Match tool sets to agent type: strategic (Read/Write/Grep), implementation (full tools), quality (sequential, heavy Bash)
• Specify execution patterns explicitly to prevent resource contention
• Test agents incrementally and version control agent files in git

Example use cases

• Generate a product-planner (strategic) agent that runs in parallel with research agents
• Create frontend-developer and backend-developer implementation agents with coordinated execution
• Produce a test-runner quality agent configured to run sequentially and access Bash for CI tasks
• Build a fullstack-coordinator to orchestrate other agents and validate integration
• Scaffold api-integration-specialist with mcp__github and mcp__playwright integrations