Build a Compound Engineering Workflow in Hermes Agent

Every Inc’s Compound Engineering methodology has gained significant traction — the open-source plugin has 17.7k GitHub stars and 1.4k forks. Built for Claude Code, Codex, Cursor, and other AI coding tools, it provides 37 skills and 51 agents organized around a core loop: Strategy → Plan → Work → Review → Compound.

But what if you’re using Hermes Agent instead? This guide shows how to implement the same compound engineering workflow using Hermes Agent’s native capabilities — skills, cron jobs, memory, and tool orchestration.

What Is Compound Engineering?

Compound Engineering is a development methodology where each feature makes the next one easier to build. The key insight is that 80% of engineering time should go to planning and review, with only 20% going to execution and documentation.

The methodology was created by Every Inc and is detailed in their Chain of Thought article. The open-source implementation is at github.com/EveryInc/compound-engineering-plugin.

The Core Loop

The Compound Engineering workflow follows a four-step cycle:

Step	Activity	Hermes Equivalent
Plan	Research requirements, explore approaches, write implementation plan	Hermes planning skills + web research
Work	Execute the plan, write tests, implement features	Hermes code execution + terminal tools
Review	Multi-perspective code review, extract lessons	Hermes code review workflow + MCP review server
Compound	Document learnings for future work	Hermes memory + knowledge skills

Mapping /ce-* Commands to Hermes

The Compound Engineering plugin provides slash commands like /ce-plan, /ce-work, and /ce-compound. Here’s how to replicate each one in Hermes Agent:

/ce-strategy → STRATEGY.md Document

The strategy document anchors all planning work. In Hermes, create a STRATEGY.md file in your project root:

# Project Strategy

## Target Problem
What problem are we solving?

## Approach
Technical direction and architecture choices

## Persona
Who uses this and what do they need?

## Metrics
How do we measure success?

## Tracks
Active work streams and dependencies

Reference this file from the project’s AGENTS.md or load it as context when starting a new feature.

/ce-brainstorm → Hermes Research + Planning Skills

The brainstorming command produces structured requirements. In Hermes, use web research tools followed by structured planning:

# ~/.hermes/skills/features/brainstorm-feature/SKILL.md
description: "Research a feature request and produce requirements"
steps:
  - Web search for similar implementations and best practices
  - Read current codebase structure and existing patterns
  - Write requirements document with success criteria
  - Review with project strategy alignment

/ce-plan → Delegated Task Plans

Compound Engineering plans include confidence checks and external research. In Hermes, use delegate_task with structured goals that include research requirements.

/ce-work → Task Execution

Work items are executed against the plan. Use Hermes terminal and file tools to implement features, with todo tracking for progress.

/ce-code-review → MCP Review Server

The Compound Engineering review pipeline uses 23 specialized reviewer agents. Hermes Agent’s MCP review server (configured via mcp_servers in config.yaml) provides structured code review. You can define multiple review personas by loading different review skills.

/ce-compound → Hermes Memory

The compounding step captures learnings. In Hermes, use the memory tool to save durable facts and patterns. Create a structured format:

Feature: [name]
Date: [date]
Decisions: [key architectural decisions]
What worked: [successful approaches]
What didn't: [failed approaches]
Follow-ups: [future improvements]

Setting Up a Knowledge Compounding Pipeline

The most important part of compound engineering is the knowledge loop. Here’s how to set it up in Hermes:

1. Create a compound knowledge skill that saves patterns to memory after completing features.
2. Configure cron jobs for periodic knowledge refresh (reviewing stale memories).
3. Load past learnings as context before starting new features by referencing memory_recall in your workflow.

Example: A Feature Development Workflow

Here’s how a complete feature development cycle looks in Hermes Agent:

# Step 1: Research and plan
# - Load STRATEGY.md for project context
# - Search web for best practices
# - Review past compound learnings
# - Write implementation plan

# Step 2: Implement
# - Execute plan tasks
# - Write tests alongside code
# - Track progress with todo

# Step 3: Review
# - Run code review via MCP server
# - Check quality gates
# - Address findings

# Step 4: Compound
# - Save learnings to memory
# - Update STRATEGY.md if needed
# - Archive old context

Getting Started

To implement compound engineering in your Hermes Agent setup:

1. Read the full Compound Engineering article for context
2. Browse the plugin source for workflow patterns
3. Create your STRATEGY.md file
4. Set up a compound knowledge skill using Hermes memory
5. Start your next feature with research and planning before execution

The methodology works with any AI coding tool — the patterns transfer, even if the specific commands don’t.