Agents

LLM-agnostic agent rules, skills, and templates for AI-assisted software engineering.

Also usable as a Claude Code plugin.

workflow.md

Human coordination guide — start here to understand how to use agents, skills, and templates together.

agents.md

Generic AI agent workflow guidelines.

agents

Agents coordinate skills and follow rules to perform specific functions.

Strategic layer

• router: Entry point for autonomous flows — routes goals and orchestrates execution
• architect: Planning specialist — decomposes features into ordered task files assigned to specialized agents, does not write code
• cofounder: Strategic cofounder — pressure-tests ideas, decisions, messaging, and artifacts through tough questions before refining them

Execution layer

• backend-engineer: Backend engineer — implements APIs, services, data models, and business logic
• frontend-engineer: Frontend engineer — implements UI from specs and task files
• fullstack-engineer: Fullstack engineer — implements features spanning frontend and backend in a single session

Quality layer

• security-engineer: Security engineer — reviews auth flows, endpoints, secrets, and permissions; produces structured findings reports
• qa-engineer: QA engineer — writes and runs Playwright E2E suites and integration tests

rules

Rules document guidelines to follow for specific contexts.

• documentation.md: [WIP] Repository and source code documentation guidelines
• react.md: [WIP] React frontend framework guidelines
• ui.md: Visual design and styling guidelines (design tokens, responsive, accessibility, animation, forms, performance)
• testing.md: [WIP] Language-agnostic testing guidelines
• python.md: [WIP] Python programming language guidelines
• fastapi.md: [WIP] FastAPI Python package guidelines
• pydantic.md: [WIP] Pydantic Python package guidelines
• TODO: typescript.md
• TODO: sqlalchemy.md
• TODO: databases.md

templates

Reusable templates for agent workflows.

• spec.md: Spec template — task description, interview, and shared context
• task.md: Task template — contract, canonical state, and session tracking

skills

Skills provide reusable workflows to perform a specific task. LLM-agnostic — any agent can follow these instructions.

• task-execute: Execute a task file by implementing the work defined in its contract
• task-plan: Plan a task by creating a spec, interviewing the human, and producing task files
• scrutinize: Pressure-test any idea, decision, artifact, or strategy through Socratic questioning and structured diagnosis
• brand-audit: Audit voice, positioning, and consistency across project artifacts

claude

Claude-specific config, patterns, etc.

Claude Code Plugin

This repo is also a Claude Code plugin. When installed, the skills above become available as slash commands (e.g., /task-plan).

Install from GitHub:

/plugin marketplace add rozelie/agents
/plugin install agents@agents

Or use locally during development:

claude --plugin-dir /path/to/agents

TODO

• Look into best practices for each agent instruction type
• Audit completed files for consistency, spelling, clarity, conciseness, duplication, etc.
• Add CONTRIBUTING.md