product-ops

This repository is deprecated. Development has moved to claude-code-engineering-leaders.

Claude Code Digital Coworkers

A collection of Claude Code plugins that give AI a well-defined operating context before it writes a single line of code. Most plugins in the broader community focus on a specific technical capability. This project takes a different angle: the plugins here focus on the context layer, giving AI the standards, constraints, and sequencing signals it needs to make better decisions about what to build and how to build it.

About This Project

This is a personal learning project. Building agents, wiring them together, and watching where they break is how I develop intuition for what works and what doesn't. The agents here are opinionated to a specific stack and working style and aren't designed for broad adoption.

If you're looking for well-maintained, general-purpose collections of Claude Code agents and skills, these are better starting points:

• wshobson/agents, a curated agent library
• affaan-m/everything-claude-code, a comprehensive Claude Code resource collection
• VoltAgent/awesome-claude-code-subagents, a community-maintained subagent catalog

What I'm Testing

Coordination and oversight, not implementation workers

The agents here are valuable because they coordinate and provide oversight — setting principles, delegating, and evaluating — not because they do implementation work that Claude already handles well on its own.

The cloud-engineering-aws plugin illustrates this with a two-tier model: devops-lead sets tool-agnostic infrastructure principles, and aws-solutions-architect translates those principles into AWS-specific architecture decisions. Each agent's Delegation section names who it consults and what it delegates, making the coordination chain explicit and inspectable.

Product owner as a guardrail against drift

One risk of delegating implementation to agents is losing sight of what actually matters. The product-owner agent is intended as a check on that: it maintains roadmap context, advises on sequencing, and pushes back when proposed work doesn't align with current priorities.

The hope is that agents coordinating with the product owner execute with better context, and that humans get clearer guidance on sequencing toward broader goals. The /write-epic, /write-story, /write-bug, and /decompose-requirement skills produce structured artifacts that land directly in GitHub Issues or Jira, keeping the backlog organized with well-refined requirements.

Benchmark harness for measurable agent quality

Agent definitions are easy to write and hard to evaluate. The benchmark harness (benchmark/benchmark.py) provides a structured way to test whether an agent's knowledge and reasoning actually hold up.

Each benchmark is a suite: a suite.yaml defining prompt templates, scoring criteria, and a dataset of question-answer pairs. The harness sends each scenario to the agent, then passes the response to an LLM judge for scoring. Adding a new suite requires no harness code changes, just a config file and a dataset.

The current suite tests aws-solutions-architect against 494 SAA-C03 exam scenarios. See benchmark/README.md for the full CLI reference and how to create new suites.

Quick Start

Add the marketplace to your Claude Code project, then install the plugins you need:

/plugin marketplace add cpliakas/claude-code-digital-coworkers
/plugin install product-owner@digital-coworkers
/plugin install cloud-engineering-aws@digital-coworkers

Plugins

product-ops

Product operations for roadmap planning, requirement authoring, and structured output for issue tracking.