Hermes Skill Forge

What Hermes Skill Forge does

Hermes Skill Forge turns repeated tasks into reusable skills. Its loop observes by searching memory and existing skills to avoid duplicates, abstracts a reusable pattern with defined inputs and outputs, writes a SKILL.md with YAML frontmatter, and tests it in a sandbox with three cases: happy path, edge case and error. If the quality score is below 0.8 it rewrites the skill and tests again. Once the score reaches 0.8, it publishes to the agentskills.io marketplace and records the skill in memory so it is not duplicated.

The README maps this to Hermes features: Memory for remembering every skill created, Skills for installing new SKILL.md files into ~/.hermes/skills/, execute_code for sandbox tests, an auto-evaluator that triggers publishing, and subagents for parallel refinement. The repository includes an Atropos RL environment and training config, a pytest suite, and three demo scenarios: web-summarizer, log-analyzer and code-reviewer. It was built for the NousResearch Show us what Hermes Agent can do hackathon.

Key features

• Detects repeated tasks and extracts a reusable pattern
• Writes SKILL.md files with YAML frontmatter into ~/.hermes/skills/
• Tests each skill with happy, edge and error cases in a Python sandbox
• Refines skills until the quality score reaches 0.8
• Atropos RL environment and training config
• Demo scenarios: web-summarizer, log-analyzer and code-reviewer

When to use it

• Studying how an agent can author and validate its own skills
• Running the demo scenarios to see the write, test, refine, publish loop
• Using the Atropos environment as a starting point for skill-writing experiments

Who it is for: Developers and researchers interested in agents that create and test their own skills.

How it fits with Hermes Agent

Built around Hermes Agent features, including Memory, Skills, execute_code, the auto-evaluator and subagents, and includes an Atropos RL environment.

How to install Hermes Skill Forge

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

pip install openai rich
set OPENROUTER_API_KEY=sk-or-...
python demo/demo_skill_forge.py --task web-summarizer

Requirements: Python with the openai and rich packages, and an OpenRouter API key.

Note: The architecture diagram describes publishing as a simulated pull request to agentskills.io, and the repository was last pushed in March 2026.

FAQ