Hermes Extend
Freshness check
Before writing or changing any Hermes config, run hermes --version and
hermes update --check, then read
https://github.com/NousResearch/hermes-agent/releases/latest. If the installed
or latest version is newer than 0.23.0, read every release note since 0.23.0
("Upgrade notes" first) and adjust. Never apply guidance from this skill to a
newer version without that check.
Decision map
| Question | Load |
|---|---|
| Write, install, lock, approve or share a Hermes skill | skills-system.md |
| Install / enable / update / pin a plugin; pick a memory provider | plugins-memory.md |
| Add an MCP server, filter its tools, gate its writes or costs | mcp.md |
| The agent must click, fill or scrape a web page | browser.md |
| Skill, script, cron script, plugin or MCP? | skill-vs-script.md |
| Ready-to-fill files | templates/hermes-skill-template.md (copy to ~/.hermes/skills/<category>/<name>/SKILL.md), templates/extend-config.yaml |
What changed recently (v0.22.0 - v0.23.0)
- Built-in skills 58 -> 44 (Word/PowerPoint/Excel merged into
microsoft-office);hermes curator lock <skill>forbids any agent rewrite, even in YOLO. - Memory providers (Honcho, Supermemory, Hindsight, Mem0, OpenViking, Holographic) moved to catalog plugins; core ships RetainDB and ByteRover.
- Plugin catalog ~600 -> 1,050+ entries, Verified tier,
hermes plugins updatefor all, opt-inplugins.auto_apply. - MCP: per-tool prices (
tool_costs) and per-chatspend_limitthat asks even in YOLO;hermes mcp servebecame themcp-serveplugin. - Browser: only
browser_execdrives the browser; the ten oldbrowser_*tools and agent-browser were removed - rewrite any skill or toolset list that names them.
Critical rules
- Never name a tool, key or plugin you have not seen in the docs, the
catalog (
hermes plugins search,hermes mcp catalog) or the source. - Third-party code is opt-in. Installing a plugin does not enable it;
plugins.enabled(or--enable) is the consent step. Read the manifest (source:, bootstrap commands) before installing a catalog MCP. - Agent-written skills are unreviewed procedures. On any profile that acts
for an owner, set
skills.write_approval: trueand lock hand-curated skills withhermes curator lock <skill>. - Untrusted MCP servers get
trust: untrustedso every write-capable call asks first;tools.includekeeps the surface minimal. - One external memory provider per profile, chosen by
memory.provider; built-inMEMORY.md/USER.mdstay on alongside it. - Deterministic, repeatable, side-effecting work is a script, not a skill (skill-vs-script.md).
Sources
- https://hermes-agent.nousresearch.com/docs/user-guide/features/skills
- https://hermes-agent.nousresearch.com/docs/developer-guide/creating-skills
- https://hermes-agent.nousresearch.com/docs/user-guide/features/plugins
- https://hermes-agent.nousresearch.com/docs/user-guide/features/memory-providers
- https://hermes-agent.nousresearch.com/docs/user-guide/features/mcp
- https://hermes-agent.nousresearch.com/docs/reference/mcp-config-reference
- https://hermes-agent.nousresearch.com/docs/user-guide/features/browser
- https://github.com/NousResearch/hermes-agent/releases/tag/v0.23.0
