hermes-internals

star 0

Class-level guide to Hermes Agent internals — authoring SKILL.md files, debugging the s6-overlay container supervision tree, fixing the TUI/CLI/gateway slash-command layer, and tracing per-model capability resolution (context window, max output tokens, pricing). Load when contributing to Hermes, debugging slash-command issues, patching model_metadata.py, or diagnosing container supervision failures.

lxgxdx By lxgxdx schedule Updated 6/13/2026

name: hermes-internals description: Class-level guide to Hermes Agent internals — authoring SKILL.md files, debugging the s6-overlay container supervision tree, fixing the TUI/CLI/gateway slash-command layer, and tracing per-model capability resolution (context window, max output tokens, pricing). Load when contributing to Hermes, debugging slash-command issues, patching model_metadata.py, or diagnosing container supervision failures. version: 1.0.0 author: Hermes Agent license: MIT metadata: hermes: tags: [hermes, hermes-internals, s6, supervisor, docker, tui, slash-commands, model-metadata, skill-authoring, context-window, gateway] related_skills: [hermes-agent, hermes-agent-skill-authoring, hermes-s6-container-supervision, debugging-hermes-tui-commands, hermes-model-metadata-debugging]


Hermes Internals — Engineering & Debugging

This umbrella covers the engineering side of Hermes Agent: the SKILL.md format, the s6-overlay container tree, the TUI/gateway slash-command layer, and per-model capability resolution. If you're using Hermes from the user's perspective, load hermes-agent instead. If you're patching, debugging, or extending Hermes internals, load this.

The four original skills folded into this umbrella (hermes-agent-skill-authoring, hermes-s6-container-supervision, debugging-hermes-tui-commands, hermes-model-metadata-debugging) cover distinct Hermes subsystems. They're preserved as references below so you can deep-dive into any one of them.

Sections

  1. SKILL.md Authoring — the original hermes-agent-skill-authoring skill. Frontmatter schema, validator, structure conventions, the peer-skill length target (8-14k chars), the references/ / templates/ / scripts/ / assets/ subdir allowlist, and worked examples of upstream-tool-adapter patterns.
  2. s6-overlay Container Supervision — the original hermes-s6-container-supervision skill. Adding/removing supervised services, debugging profile gateways that won't start, understanding the Architecture B main-program pattern (/opt/hermes/docker/main-wrapper.sh + leading-dash args).
  3. TUI Slash Command Debugging — the original debugging-hermes-tui-commands skill. The 3-layer architecture (Python registry → tui_gateway JSON-RPC → Ink/TypeScript UI), why commands work in CLI but not TUI, the autocomplete miss, the config-persistence/UI-update race.
  4. Model Metadata Debugging — the original hermes-model-metadata-debugging skill. Where the 200K/32K/1M context window comes from, how to override it, the agent/model_metadata.py and agent/models_dev.py tables, the layering order (override → table → models.dev → fallback).
  5. MiniMax-M3 Case Study — the actual 2026-06-01 session trace of a metadata resolution problem, showing the search commands and the layering findings applied to a real model.
  6. Upstream Tool Adapter Pattern — worked example: how to wrap a third-party CLI tool as a Hermes skill (Hermes sessions → SkillOpt-Sleep pattern).
  7. Skill Evolution from History — how to mine session history for skill patterns and persist them.

When to Load This Umbrella

  • Writing or reviewing a new SKILL.md (or fixing an existing one)
  • Debugging a Hermes Docker container (s6 service won't start, gateway won't boot)
  • Diagnosing why a slash command works in CLI but not in TUI (or vice versa)
  • Resolving a model metadata disagreement (Hermes says 200K, the provider says 1M)
  • Contributing a tool, plugin, or platform adapter
  • Auditing a Hermes installation for internal consistency

Scripts

  • scripts/trace-context-window.py — runs the Hermes context-window resolver end-to-end and prints which layer (override / table / models.dev / fallback) supplied the number. Use this when debugging a new model — saves you from re-grepping the source.

Decision: Load hermes-agent or This Umbrella?

Task Skill to load
Configure Hermes (CLI, profile, gateway, voice) hermes-agent
Use a Hermes feature (cron, kanban, webhook) hermes-agent
Write a new SKILL.md this umbrella (§1)
Debug a container service this umbrella (§2)
Fix a TUI slash command this umbrella (§3)
Resolve model metadata this umbrella (§4)
Spawn another Hermes instance hermes-agent
Troubleshoot the Kanban board hermes-kanban

Related Skills

  • hermes-agent — the user-facing umbrella (CLI, setup, gateway, profiles). Load this for "how do I use Hermes?" questions.
  • hermes-kanban — the Kanban multi-agent system (worker/orchestrator/codex-lane).
  • hermes-agent-skill-authoring, hermes-s6-container-supervision, debugging-hermes-tui-commands, hermes-model-metadata-debugging — the four original skills that were merged into this umbrella. Now archived; their full content lives in the references/ above.
Install via CLI
npx skills add https://github.com/lxgxdx/hermes-skills --skill hermes-internals
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator