Stop Re-Explaining Your Work to Hermes

If you have to explain the same job again every time you open Hermes, put the job in a small folder first.

You do not need a huge system. Start with one repeating review that is safe to prepare as a draft and still needs your approval before anything happens. The goal is simple: when you open a new chat, the work, the boundary, and the latest decision are already findable.

## Pick a review that is safe to draft

Choose something you already do more than once. Good first examples include:

- reviewing a few support requests and preparing reply drafts

- checking a weekly project list and flagging what needs attention

- gathering a small set of updates into a decision brief

Do not begin with payments, account changes, sending messages, deleting files, or anything where a bad first draft could hurt someone. Your first workflow should prepare information, not take action.

## Make one folder with three files

Create a folder named for the job. Inside it, make these three plain-text files:

```text

weekly-project-review/

├── PURPOSE.md

├── CONTEXT.md

└── decisions.md

```

### 1. `PURPOSE.md`

Write what the review is for, what comes in, and what should come out.

```markdown

# Weekly Project Review

## Purpose

Prepare a short draft of this week's priorities for my review.

## Inputs

- Current project notes

- Last review's decisions

## Output

- A draft priority list with open questions

## Human check

I review the draft before any task, message, calendar, or file change happens.

```

### 2. `CONTEXT.md`

Write the facts Hermes needs every time. Keep it short and current.

```markdown

# Context

- We work on three active client projects.

- Jordan makes final priority decisions.

- Do not contact clients or change project records.

- Flag missing information instead of guessing.

```

### 3. `decisions.md`

Add the latest choices and unresolved questions. Date each entry.

```markdown

# Decisions

## 2026-09-11

- Finish the proposal draft before starting the website refresh.

- Open question: Is the client meeting still scheduled for Thursday?

```

Those files are the handoff. You can open them. A teammate can read them. A new Hermes chat can use them without you rebuilding the whole explanation from memory.

## Ask Hermes for a draft, not an action

Open Hermes in that folder and use a prompt like this:

> Read `PURPOSE.md`, `CONTEXT.md`, and `decisions.md`. Prepare this week's review draft. List missing information and open questions clearly. Do not send messages, change records, or take any external action.

Read the result. Edit the files if the result missed something. Then run it again. The folder gets better because the important context lives where you can see it.

## Test it with a clean chat

The useful test is not whether one conversation produces a good answer. It is whether a fresh chat can find the job without you re-explaining it.

Start a new chat. Point Hermes to the folder. Ask it to say:

- what the job is

- what it is allowed to do

- what needs a human decision

- what was decided most recently

If those answers are clear, you have built a working starting point.

## When this approach fits, and when it does not

Folders work well for repeatable work with a clear order and a person who reviews the output. They are a poor fit for high-volume, real-time work that needs many agents coordinating at once. For a one-off task, a saved prompt may be enough.

The point is not to build more process. It is to stop keeping the process trapped in your head or in an old chat window.

## What sits underneath this approach

This draft uses the practical idea that files and folders can make sequential, human-reviewed agent work easier to inspect and edit. The formal method is Interpretable Context Methodology, described by Jake Van Clief and David McDermott in their paper, [*Interpretable Context Methodology: Folder Structure as Agent Architecture*](https://arxiv.org/abs/2603.16021). AgentsAsFolders.ai is not affiliated with or endorsed by the authors.

For Hermes basics, start by proving a normal chat works before adding more layers such as bots, schedules, or automations. See the official [Hermes Agent Quickstart](https://hermes-agent.nousresearch.com/docs/getting-started/quickstart).

18 comments

Stop Re-Explaining Your Work to Hermes