Skip to content

groktopus/loop-designer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

loop-designer

Autonomous agent loop design in 10 steps.

Version

A Hermes Agent skill that distills Alex (@de1lymoon)'s 10-step roadmap into concrete Hermes Agent invocations — from /goal with goal_judge to unattended cron loops with memory, verification, and fail-safe guardrails.

This is NOT about better prompting or model selection. The model stays the same — what improves is the infrastructure wrapped around it.

What it covers

Step Concept Mechanism
1 Agent is a loop /goal loops via goal_judge
2 Harness first Dedicated profile with tool-config pruning
3 System improves, not model Memory + skills compound across runs
4 Independent grader goal_judge + kanban tasks with acceptance criteria
5 Maker/checker split delegate_task verifier subagent
6 Timer, then cloud profile-created cronjob seeds kanban on 60s tick
7 Compose with workflows delegate_task batch + kanban deps
8 Memory State log + durable memory + skills
9 Distill into skills skill_manage patch-on-fail lifecycle
10 Fail safe Restricted profile + enabled_toolsets + repeat cap

Prerequisites

  • Hermes Agent with /goal and /subgoal commands
  • goal_judge auxiliary model configured
  • Kanban dispatcher, cronjob, skill_manage, and memory tools
  • A dedicated Hermes profile for running the loop

Installation

Clone the repo into your Hermes skills directory:

git clone https://github.com/groktopus/loop-designer.git ~/.hermes/skills/thinking/loop-designer

Verify it's installed:

hermes skills list | grep loop-designer

Quick start

Once installed, load the skill with the slash command:

/loop-designer

Or reference it naturally in conversation — Hermes discovers and loads skills automatically when their triggers match.

Then work through the 10 steps sequentially — each step in SKILL.md translates one move from the article into a concrete Hermes configuration or tool invocation. The canonical output is a cron job created from a dedicated profile context, with goal_judge as the grader, explicit completion contracts, a repeat cap, append-only state, and a skill-update lifecycle.

Documentation

The full methodology — all 10 steps with configuration examples and pitfalls — lives in SKILL.md.

License

MIT © 2026 groktopus — see LICENSE.

About

Autonomous agent loop design in 10 steps — from /goal to unattended cron loops with memory, verification, and fail-safe guardrails

Resources

License

Stars

1 star

Watchers

1 watching

Forks

Packages

 
 
 

Contributors