DEV Community

klauski24
klauski24

Posted on

I Built Hermes Skill Gardener: An Agent That Notices Repeated Work

#hermesagentchallenge #devchallenge #agents

Hermes Agent Challenge Submission: Build With Hermes Agent

I Built Hermes Skill Gardener: An Agent That Notices Repeated Work

Hermes Skill Gardener detects repeated workflows, asks for confirmation, runs the saved checklist, and learns a better rule for next time.

This is a submission for the Hermes Agent Challenge: Build With Hermes Agent

Repository: https://github.com/klauski24/hermes-skill-gardener

Hermes Skill Package: hermes-skill/skill-gardener/SKILL.md

What I Built

Hermes Skill Gardener is a small agentic workflow that detects when a user is
about to repeat a task they have done before.

Instead of making the user explain the same process again, it finds the closest
stored skill, asks for confirmation, runs the workflow, and then records new
lessons for next time.

The guiding idea:

You should not have to teach your agent the same workflow twice.

Why I Built It

Many AI demos are good at answering one prompt. Many automation tools are good
at running one fixed recipe. I wanted to explore the space between those two
ideas: an agent that can turn repeated work into reusable behavior.

Hermes Agent is a strong fit for this because its core story is not just tool
use. It also has memory, skills, persistence, and a learning loop.

How It Works

  1. The user enters a new task.
  2. The agent compares that task with stored skills.
  3. If a similar workflow exists, the agent asks whether to run it.
  4. If confirmed, the agent generates the execution checklist.
  5. After the run, the agent stores a new lesson.

What Is Included

  • A dependency-free Python CLI.
  • A local web demo.
  • A persisted skill memory file.
  • Markdown run briefs.
  • A Hermes-compatible SKILL.md export.
  • Unit tests for matching and skill packaging.
  • Ranked match alternatives with low-confidence handling.
  • A browser-based lesson form for teaching the agent after a run.

Example 

User: I need to prepare a DEV challenge submission for a new AI agent project.

Hermes Skill Gardener:
This looks like `dev-challenge-submission` with 40% similarity.
Should I run that workflow?
Enter fullscreen mode Exit fullscreen mode

When confirmed, it creates a run brief containing the workflow, past lessons,
and suggested next actions.

Demo

The fastest demo is:

python skill_gardener.py detect "I need to prepare a DEV challenge submission for a new AI agent project"
python skill_gardener.py run "I need to prepare a DEV challenge submission for a new AI agent project" --yes
python skill_gardener.py serve
Enter fullscreen mode Exit fullscreen mode

There is also a production-oriented version:

python production_app.py
Enter fullscreen mode Exit fullscreen mode

It adds registration, login, SQLite persistence, per-user skill memory, and a
Dockerfile for deployment.

Code

Repository: https://github.com/klauski24/hermes-skill-gardener

My Tech Stack

  • Python standard library
  • SQLite
  • Vanilla HTML, CSS, and JavaScript
  • Docker
  • Hermes-compatible SKILL.md package

How I Used Hermes Agent

The project is built around the Hermes idea of persistent skills. It exports a
Hermes-compatible package at hermes-skill/skill-gardener/SKILL.md, with helper
scripts and a starter skill memory file.

In a Hermes session, the workflow is:

  1. Detect whether the user is repeating a known workflow.
  2. Ask for confirmation if the match is strong enough.
  3. Run the saved workflow checklist.
  4. Execute the checklist with the tools available in Hermes.
  5. Save a new lesson after the run so the next execution improves.

Why This Is Agentic

  • It uses memory to recognize repeated work.
  • It plans with reusable workflow steps.
  • It keeps a human in the loop before action.
  • It improves its own future behavior by saving lessons.
  • It becomes more useful the more it is used.

Judging Criteria Mapping

  • Effective use of Hermes capabilities: designed around skills, memory, and reusable procedural workflows.
  • Technical implementation: CLI, web demo, tests, safe JSON persistence, and a generated Hermes skill package.
  • Creativity: focuses on the moment before automation, where the agent detects repeated intent and asks whether to reuse learned behavior.
  • Usability: one-command demo, readable run briefs, and a browser interface.

How To Run 

python skill_gardener.py list
python skill_gardener.py detect "I need to prepare a DEV challenge submission for a new AI agent project"
python skill_gardener.py run "I need to prepare a DEV challenge submission for a new AI agent project" --yes
python skill_gardener.py serve
Enter fullscreen mode Exit fullscreen mode

The full installation guide is in INSTALL.md.

What I Learned

The most interesting agent workflows are not always the flashy ones. Sometimes
the valuable part is noticing that the user is about to do something familiar
and offering to handle the boring structure around it.

That is where an agent starts to feel less like a chatbot and more like a
personal operating system.

Links

Top comments (0)