Hermes Agent Engineering Standard

What Hermes Agent Engineering Standard does

Hermes Agent Engineering Standard sets out rules for turning an incompletely defined development goal into a software change that can be reviewed, implemented, verified, merged and traced. It is not tied to a programming language, framework, database, code-hosting method or AI model, and the process can be scaled to task risk, whether one person works with AI help or several agents work in parallel. Humans decide what the product should do, while agents with clearly defined responsibilities carry out investigation, design, implementation and verification.

The README lists recurring risks and pairs each with a control: fact drift, false completion, over-reliance on context packages, self-review and self-approval, unsafe execution against real resources such as databases, irrecoverable state after interruption, and design deviation hidden by green tests. Current code, configuration and runtime evidence take precedence over documents. Verification Levels 0 to 3 separate local correctness, integration correctness, system correctness and correctness in the real environment. Responsibilities for requirements and design, implementation, review, integration and verification are kept apart.

Key features

• Verification Levels 0 to 3, from local correctness to correctness in the real environment
• Separate responsibilities for requirements and design, implementation, review, integration and verification
• Rule that current code, configuration and runtime evidence outrank documents and context packages
• Process chosen by task risk, for a single developer or parallel agents
• Table of common failure modes, each tied to a control mechanism

When to use it

• Setting review and approval rules for several coding agents working in parallel
• Deciding when an AI-made change counts as verified rather than merely compiled
• Drafting team policy on what agents may do to databases and other real resources

Who it is for: Engineering leads and solo developers who use AI coding agents and want written rules for scope, review and verification.

How it fits with Hermes Agent

The repository name refers to Hermes Agent, but the README text describes a tool-agnostic standard that is not tied to any AI model.

Note: The README describes a general, model-agnostic standard and does not show Hermes-specific files or commands.

FAQ