Repolode
Submit
Yonkoo11

Yonkoo11/hermes-dojo

A self-improvement add-on for Hermes Agent that analyzes session failures, patches weak skills, runs GEPA evolution, and reports progress.

View on GitHub
Stars
187
Forks
17
Open issues
0
Last commit
5 months ago

What is it?

Hermes Dojo is a self-improvement system for Hermes Agent. It reads the agent's session logs from state.db, flags failures such as tool errors, retry loops and user corrections, and tracks success rates for each skill.

The system ranks weaknesses by root cause and then acts on them. It patches existing skills that fail, creates new skills when users repeatedly ask for something none exists for, and runs GEPA self-evolution on weak skills. Results are stored as daily metrics so the learning curve can be viewed over time.

Dojo installs as a Hermes skill and exposes commands such as /dojo analyze, /dojo improve, /dojo report and /dojo history. An overnight cron mode runs the full cycle and can deliver the morning report to Telegram or the CLI. It requires Hermes Agent v0.2.0+, Python 3.10+, and the hermes-agent-self-evolution package for GEPA.

Key features

  • 01Parses agent session logs to detect tool errors, retry loops and user corrections.
  • 02Tracks per-skill success rates and ranks the weakest skills by root cause.
  • 03Patches failing skills with targeted error handling.
  • 04Creates new skills from repeated unmet user requests.
  • 05Runs GEPA self-evolution on weak skills and stores daily metrics for a learning curve.
  • 06Sends reports to CLI or Telegram, with an optional overnight cron cycle.

Use cases

  • Keeping a Hermes Agent's skills reliable as tool APIs change or time out.
  • Getting a daily summary of where the agent's skills are failing.
  • Automatically drafting new skills for tasks a user repeatedly requests.
  • Measuring whether agent performance improves over weeks.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/Yonkoo11/hermes-dojo.git
  2. 2

    Enter the directory and run the installer

    cd hermes-dojo && ./install.sh
  3. 3

    Seed demo data (optional)

    cd ~/.hermes/skills/hermes-dojo/scripts && python3 seed_demo_data.py --days 7
  4. 4

    Run the full pipeline demo

    python3 demo.py --reset

Things to know

  • Requires Hermes Agent v0.2.0+ and Python 3.10+.
  • Self-evolution (GEPA) also needs the separate hermes-agent-self-evolution project.

Topics

#ai-agents#self-improvement#hermes-agent#llm#skills#observability#automation

FAQ

What is hermes-dojo?
A self-improvement add-on for Hermes Agent that analyzes session failures, patches weak skills, runs GEPA evolution, and reports progress. Hermes Dojo is a self-improvement system for Hermes Agent. It reads the agent's session logs from state.db, flags failures such as tool errors, retry loops and user corrections, and tracks success rates for each skill.
Is hermes-dojo free to use? What license does it have?
hermes-dojo is open source under the MIT license. Check the license terms before using it in commercial products.
Is hermes-dojo still maintained?
Updated within 6 months. The last commit to hermes-dojo was on Jun 6, 2026.
What can I use hermes-dojo for?
Keeping a Hermes Agent's skills reliable as tool APIs change or time out. Getting a daily summary of where the agent's skills are failing. Automatically drafting new skills for tasks a user repeatedly requests. Measuring whether agent performance improves over weeks.
How do I install hermes-dojo?
Clone the repository: git clone https://github.com/Yonkoo11/hermes-dojo.git Enter the directory and run the installer: cd hermes-dojo && ./install.sh Seed demo data (optional): cd ~/.hermes/skills/hermes-dojo/scripts && python3 seed_demo_data.py --days 7 Run the full pipeline demo: python3 demo.py --reset
Where can I get hermes-dojo?
The source code and installation instructions are on GitHub: https://github.com/Yonkoo11/hermes-dojo

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects