Hermes Agent Skills
Skills are the vehicle for Hermes's "self-improvement": they distill problems you've solved into reusable, shareable units of capability that get stronger with use.
What Is a Skill?
A skill is a reusable procedure/capability packaged in a standard format under ~/.hermes/skills/. Unlike a one-off tool call, a skill is named, saved, and searchable:
- Reusable: next time a similar problem appears, just invoke it
- Shareable: follows the agentskills.io open standard for export/import
- Auto-generated: after solving a complex task, Hermes writes the method into a skill document
- Self-improving: refines more reliable paths through use
SKILL.md Format
Skills are described in SKILL.md / .skill files following the open standard. A skill typically contains:
- Name and description: used to judge relevance during retrieval
- Trigger / applicability
- Steps: which tools to call, in what order
- (optional) Script: an executable definition in Python/YAML
Because it follows the agentskills.io open standard, community skills are reusable across agents — a key value of the Hermes skill ecosystem.
Three Ways to Get Skills
Using Skills in Conversation
Browse and run skills with slash commands:
For example:
Hermes can also invoke a fitting skill automatically when it judges it relevant.
Auto-Generated Skills: The Heart of Self-Improvement
This is what sets Hermes apart from ordinary assistants:
Over time, your Hermes accumulates a skill library tailored to your workflow — the essence of a "growing agent."
Writing Your Own Skills
You can also create skill files by hand under ~/.hermes/skills/. Recommendations:
- Clear name, precise description: retrieval relies on the description
- Single responsibility: one skill does one thing, easy to compose
- Composable: build complex flows from small skills
- Verify with
/skillsthat it can be found and run
Practical Tips
- Prioritize distilling things you ask Hermes to do repeatedly into skills
- Use community skills to avoid reinventing the wheel
- Periodically review your skill library; prune stale/low-quality ones (keep signal high)
- Teams can share one skill set for a consistent workflow
Next Steps
- Memory System — skills handle "doing," memory handles "remembering"
- Slash Commands —
/skillsand related commands - Multi-Agent — run skills in parallel across sub-agents