hermes

name	hermes
description	Expert at understanding and working with Hermes Agent — its memory system, skills, cron jobs, tools, and behavioral conventions. Use when configuring Hermes, writing skills for it, debugging its behavior, or understanding what it can and cannot do.
trigger	how does hermes work, hermes memory, hermes skills, hermes tools, hermes configuration, hermes capabilities, what can hermes do

Hermes Agent — Self-Knowledge

Hermes is an autonomous AI agent by Nous Research that persists across sessions, builds memory of users, and executes scheduled tasks.

Core Architecture

• Memory: Dual-store — memory (agent notes) + user (user profile). Persists via Honcho between sessions. Injected as header into every turn.
• Skills: Procedural memory at ~/.hermes/skills/{category}/{skill}/SKILL.md. Auto-loaded when relevant. Converted from experience after complex tasks.
• Cron Jobs: Scheduled tasks at ~/.hermes/cron/jobs.json. Run in isolated sessions. Output saved to ~/.hermes/cron/output/{job_id}/{timestamp}.md.
• Tools: 40+ built-in tools + MCP servers. Discovery at startup, registered in tool registry.

Memory System

══════════════════════════════════════════════
MEMORY (your personal notes) [85% — 1,890/2,200 chars]
══════════════════════════════════════════════
Entries separated by § (section sign)
Entries can be multiline
Capacity % shown so agent knows when to prune

Instruction Capture System: When user states a preference or correction, ask "Save to instructions?", synthesize to ~/.hermes/instructions.md (not a literal copy — a refined version).

Memory commands:

• memory_add — save durable facts
• honcho_conclude — write peer profile conclusions
• session_search — recall past sessions

Skills System

Skills trigger automatically based on keywords in the user's request. After a complex task (5+ tool calls), offer to save the approach as a skill.

Skill structure:

~/.hermes/skills/{category}/{skill-name}/
├── SKILL.md              # Required: frontmatter + instructions
├── references/           # Optional: detailed docs
├── scripts/              # Optional: executable code
└── assets/               # Optional: templates, data

Creating skills:

1. Complex task succeeded → offer to save
2. Errors overcome → update existing skill
3. User corrected approach → patch immediately
4. Use skill_manage with action=create/patch/edit/delete

Cron Jobs

Jobs are defined in ~/.hermes/cron/jobs.json with fields: name, prompt, schedule, deliver, enabled.

Schedule formats: cron expressions (0 9 * * *), intervals (30m, every 2h), ISO timestamps.

Delivery targets: local (save to output dir), origin (reply to source), or platform (telegram, discord, whatsapp, signal, etc.).

Key issue: The scheduler's tick() function must be called by the gateway every 60 seconds. If the gateway process doesn't import/run the scheduler, jobs never auto-fire — only manual run triggers work.

Tool Categories

Category	Tools
Terminal	terminal, process — run shell commands, manage background jobs
Files	read_file, write_file, patch, search_files
Browser	browser_navigate, browser_snapshot, browser_click, browser_vision
Web	web_search, web_extract
Cron	cronjob — create, list, run, pause, remove scheduled jobs
Memory	memory, honcho_conclude, session_search
Skills	skill_manage, skill_view, skills_list
Code	execute_code — Python script with hermes_tools
Delegation	delegate_task — spawn subagents for isolated parallel work
Vision	vision_analyze — AI image analysis
TTS	text_to_speech — convert text to audio

Key Conventions

• Communication style: Concise, direct. Code speaks louder than words.
• Modes: User prefers yolo for creative work, precise for health/critical tasks.
• Context management: User separates subjects into different chats to avoid context bleeding.
• Memory priority: User preferences and corrections > environment facts > procedural knowledge.
• No repeat work: Check session_search before asking user to repeat themselves.

Documentation

Reference docs cached locally from upstream:

Getting Started

• Installation - Setup and install
• Quickstart - First steps
• Learning Path - Guided learning

User Guide

• CLI - Command-line interface
• Configuration - Config options
• Messaging - Messaging platforms
• Profiles - Running multiple agents
• Security - Security model

Features

• Tools - Built-in tools
• Memory - Memory system
• Skills - Skills system
• MCP - MCP integration
• Voice Mode - Voice interaction
• Personality - Personality customization
• Context Files - Context file system
• Cron - Scheduled tasks

Guides

• MCP Guide - Using MCP with Hermes
• Voice Guide - Voice mode guide
• Tips - Usage tips

Reference

• CLI Commands - Full command reference
• FAQ - Frequently asked questions
• Architecture - System architecture
• Source Reference - Repo structure, all tools, toolsets, platforms, config sections, env vars
• Config Reference - Full cli-config.yaml.example
• Env Reference - Full .env.example

Upstream

• Docs: https://hermes-agent.nousresearch.com/docs
• Repo: https://github.com/NousResearch/hermes-agent
• Skills spec: https://agentskills.io