Hermes Prompt Optimizer Workflow

A Hermes Agent skill that turns project-related requests into clear, approval-based execution prompts before the agent edits, changes, or builds anything.

Built for NousResearch Hermes Agent, an open-source AI agent with a skill system.

Install with skills.sh

To install through the open agent skills CLI, run:

npx skills add BRANDZ0/hermes-prompt-optimizer-workflow

View the indexed skill page here:

https://skills.sh/brandz0/hermes-prompt-optimizer-workflow/hermes-prompt-optimizer-workflow

This repo uses the standard skill folder layout:

skills/hermes-prompt-optimizer-workflow/SKILL.md

Hermes terminal install option

If you prefer installing it as a Hermes skill through the Hermes CLI, run:

hermes skills install BRANDZ0/hermes-prompt-optimizer-workflow --now

Then in Hermes, say:

Use the hermes-prompt-optimizer-workflow skill for this project. For every project-related task, improve the prompt first, ask for missing context if needed, show me the final prompt, and wait for approval before executing.

If your current session does not pick it up, start fresh:

/reset

Easy paste setup

If you do not want to use terminal commands, paste this into Hermes:

Read this repo and follow the Hermes Prompt Optimizer Workflow for all project-related tasks:
https://github.com/BRANDZ0/hermes-prompt-optimizer-workflow

Use skills/hermes-prompt-optimizer-workflow/SKILL.md as the main instruction file.

That is enough for most users if Hermes can read the repo link.

Update later

When this repo gets improved, update the skill with:

hermes skills check
hermes skills update hermes-prompt-optimizer-workflow

If Hermes warns because this is a community GitHub skill, review the repo first. If you trust it, install with:

hermes skills install BRANDZ0/hermes-prompt-optimizer-workflow --now --force

What Hermes will do

For every project-related task, Hermes should:

1. Identify the task type.
2. Ask for missing context, screenshots, files, logs, or examples.
3. Rewrite the request into a stronger execution prompt.
4. Show the improved prompt to the user.
5. Wait for approval.
6. Execute only after the user approves.

What this does

Instead of letting Hermes immediately execute vague requests, this workflow makes Hermes pause and improve the task first.

Example request:

fix this mobile ui the buttons look funny

Hermes should not start editing right away.

It should first create a better plan/prompt, ask for anything missing, show the final prompt, and wait for approval.

Workflow chart

flowchart TD
    A[User gives project-related task] --> B[Hermes identifies task type]
    B --> C[Hermes checks what context is missing]
    C --> D{Is more context needed?}
    D -- Yes --> E[Ask for screenshots, files, logs, examples, or repo details]
    E --> F[Create improved execution prompt]
    D -- No --> F[Create improved execution prompt]
    F --> G[Show final prompt to user]
    G --> H{User approves?}
    H -- Yes --> I[Execute the task]
    H -- No --> J[Revise the prompt]
    J --> F
    I --> K[Summarize changes and what to test]

Loading

When Hermes should use this

Use this workflow for every project-related task:

Task type	Examples
UI and design	Layout fixes, mobile responsiveness, spacing, colors, component polish
Bugs	Broken login, console errors, failed API calls, bad state behavior
Features	New pages, buttons, flows, forms, dashboards, settings
Refactors	File cleanup, component splitting, safer structure, reusable logic
Docs	README files, setup guides, usage docs, repo instructions
Deployment	Vercel, Railway, VPS, Netlify, build errors, environment variables
Repo tasks	Branch work, cleanup, config files, project organization
Prompt writing	Agent prompts, coding prompts, system instructions, task templates

Core rule

Hermes should not execute project-related tasks immediately.

Hermes must first optimize the request, ask for missing context when helpful, show the improved prompt, and wait for approval.

Main files

File	Purpose
skills/hermes-prompt-optimizer-workflow/SKILL.md	Main skill entrypoint with metadata
skills/hermes-prompt-optimizer-workflow/references/	Skill reference files for context questions, examples, and deeper rules
COPY_THIS_TO_HERMES.md	Simple paste-in instructions for Hermes
PROMPT_OPTIMIZER.md	Root copy of the full workflow rules
CONTEXT_QUESTIONS.md	Root copy of context questions
EXAMPLES.md	Root copy of examples
CHANGELOG.md	Recent workflow changes and update notes

Keeping the workflow updated

This skill tells Hermes to re-read the latest repo files at the start of new chats, project sessions, or repo sessions.

The latest behavior is defined in:

1. skills/hermes-prompt-optimizer-workflow/SKILL.md
2. skills/hermes-prompt-optimizer-workflow/references/
3. README.md
4. CHANGELOG.md

Expected behavior

When installed correctly, Hermes should respond to project tasks like this:

I will optimize this task before executing.

Task type: UI / mobile responsiveness

I need one screenshot of the current mobile UI and the screen size where it looks wrong. After that, I will create the final execution prompt for your approval before making changes.

Then after context is provided, Hermes should show something like:

Final execution prompt:

Inspect the mobile button layout and fix spacing, sizing, alignment, and responsiveness issues. Preserve existing functionality. Keep changes small and focused. Test common mobile widths. Do not redesign unrelated sections.

Approve this prompt before I execute?

Recommended approval words

approved
approve and run
yes execute
run it

Recommended default behavior

Hermes should always:

• preserve existing functionality
• avoid large rewrites unless needed
• ask for screenshots on UI tasks
• ask for logs on bug or deployment tasks
• inspect relevant files before repo edits
• keep changes focused
• summarize what changed
• tell the user what to test
• avoid exposing secrets, tokens, keys, or private data

Not a prompt marketplace

This repository is not a prompt marketplace and does not copy private prompts from any service.

It is an open workflow pattern for making Hermes Agent pause, gather context, improve the task, request approval, and then execute safely.

Best use case

This works best for people who use Hermes Agent or coding agents and want more control before the agent makes changes.

Ideal for:

• Hermes Agent skills
• Hermes project workflows
• Claude Code style workflows
• Cursor-style workflows
• Codex-style workflows
• repo-specific AI instructions
• team prompt standards
• reusable AI operating procedures

License note

Use and modify this workflow for your own projects. Do not include secrets, private repo details, or personal information when sharing it publicly.