hermes-agent

name	hermes-agent
description	Hermes Agent integration for CKS — project scaffolding, AGENTS.md generation, skill mapping, and Hermes-native templates. Use when the user asks for Hermes support, multi-harness scaffolding, or when generating AGENTS.md. Maps Hermes concepts (skills, memory, profiles, MCP tools, cron jobs) to CKS equivalents.
allowed-tools	Read, Write, Edit, Glob, Grep, Bash
model	sonnet

Hermes Agent Integration for CKS

Teaches CKS about Hermes Agent so generated projects work across both Claude Code and Hermes. Key insight: a "modern AI harness" project has AGENTS.md for Hermes AND CLAUDE.md for Claude Code — sometimes alongside .cursorrules for Cursor.

What Hermes Agent Is

Hermes is an AI agent platform by Nous Research. It runs as a CLI, TUI, or background daemon. Key concepts:

Hermes Concept	CKS Equivalent	Notes
Skills (.hermes/skills/)	CKS skills (skills/)	Procedural knowledge loaded on demand
AGENTS.md	CLAUDE.md	Standing orders for the agent session
Memory (persistent store)	.learnings/	Cross-session facts vs session artifacts
Profiles (per-project config)	.prd/ config	Hermes isolates by profile, CKS by project
MCP tools	CKS tools/	Model Context Protocol servers
Cron jobs (scheduled)	scheduled-agents/	Recurring agent tasks
Toolsets	allowed-tools frontmatter	Named capability groups

Hermes Project Architecture

When CKS scaffolds a project intended to work with Hermes, generate this structure:

project/
├── AGENTS.md                 ← Standing orders for Hermes (required)
├── CLAUDE.md                 ← Standing orders for Claude Code (optional)
├── .hermes/
│   ├── config.yaml           ← Hermes profile config
│   ├── rules/                ← Glob-scoped guardrails (same concept as .claude/rules/)
│   ├── plans/                ← Plan files (like CKS plan files)
│   └── skills/               ← Project-local Hermes skills
├── .claude/                  ← Only if Claude Code is also needed
├── skills/                   ← Shared CKS skills (loaded by both)
├── src/
├── tests/
├── package.json / pyproject.toml
└── README.md

Rule: When to Generate Which Files

User Goal	Generate
"I use Hermes"	AGENTS.md + .hermes/
"I use Claude Code"	CLAUDE.md + .claude/
"I use both"	AGENTS.md + CLAUDE.md + shared skills/
"I use Cursor/Codex/Copilot"	.cursorrules or appropriate config

AGENTS.md Template (Hermes Standing Orders)

Use this template when generating AGENTS.md:

# AGENTS.md — Standing Orders

## Auto-Load Skills
Always load these skills silently at session start:
- {skill-dependencies}

## Pre-Ship Invariant
Before ANY deploy, release, or production-adjacent action:
1. Run the security gate (defined below)
2. If CRITICAL findings exist, STOP and demand fix
3. Document in .hermes/decisions/

## Maturity Stage
Current stage: {maturity-stage}

Behaviors per stage:
- **Prototype**: Speed over perfection. Never commit secrets, never skip auth.
- **Pilot**: Full security audit before any user-facing change.
- **Candidate**: All gates must pass.
- **Production**: No deploy without passing CI.

## Proactive Behaviors
- Flag security issues immediately
- Suggest tests BEFORE implementing
- Document decisions in .hermes/decisions/
- Never say "that's a good idea" to disabling security

## Memory Sync
At session end, capture:
- Branch and commits
- Security findings (even if resolved)
- Architecture decisions made

Hermes config.yaml Template

# Hermes profile config
provider: openrouter  # or anthropic, openai
model: deepseek/deepseek-v4-flash  # or claude-sonnet-4, etc.

# Auto-load skills at session start
skills:
  - ciso-security
  - soul-framework

# MCP server configuration
mcp_servers:
  - package: n8n
    args:
      baseUrl: ...  # n8n webhook URL
  # Add project-specific MCP servers here

# Tool configuration
tools:
  enabled_toolsets:
    - terminal
    - file
    - web
    - browser

Hermes Skills vs CKS Skills

Hermes skills live under ~/.hermes/skills/<name>/ or project-local .hermes/skills/<name>/. They use the same YAML frontmatter + markdown structure as CKS skills.

Key difference: Hermes skills are auto-loaded by name via AGENTS.md, not by plugin discovery. CKS skills are loaded by agent frontmatter.

When generating skills during /cks:kickstart, create them in .hermes/skills/ if the user uses Hermes, and skills/ if they use Claude Code.

Hermes-Compatible MCP Server Templates

When the project needs MCP servers (the modern pattern for AI tool integration), generate an MCP config section and reference the cks-stack stack-layers:

# In .hermes/config.yaml or project MCP server list
mcp_servers:
  - package: filesystem
    args:
      allowed_directories: ["./"]
  - package: github
    args: {}
  # Custom per-project MCP servers

Verification

• AGENTS.md has standing orders matching CKS project lifecycle
• .hermes/config.yaml has correct provider and model
• Skills referenced in AGENTS.md exist in skills/ directory
• MCP server config matches project requirements
• Security gates defined for pre-ship invariant