Explain and draft.
Use for first-pass brainstorming, kid-simple explanations, prompt cleanup, and turning messy notes into a plan.
20-minute quick start.
Do this before connecting apps, buying tools, or asking an agent to “run the business.” The goal is one safe repeatable workflow.
Step 01Pick one boring repeated job: weekly review, follow-up draft, content outline, inbox summary, appointment prep, report builder.one job
Step 02Write the approved inputs: local notes, public URLs, worksheet fields, screenshots you are allowed to use.source lock
Step 03Ask ChatGPT or Claude to turn the job into plain-English steps.plan
Step 04Ask Codex/OpenClaw-style agent to create local files or a page only after the plan is clear.build
Step 05Use Hermes to save the repeated procedure as a skill or schedule a draft-only review report.repeat
Step 06Human approves any send, publish, account, payment, record, or deployment action.gate
Which model should I use?
Simple routing for people who do not care about model wars. Pick by job, not hype.
Review and reason.
Use for long documents, nuanced voice, strategy, safety review, and high-stakes judgment before publishing.
Build and test.
Use for code, local pages, scripts, QA, refactors, and verified tool output inside a project folder.
Multimodal thinking.
Use when screenshots, files, documents, or large-context review are the center of the work.
Agent execution.
Use as a local task agent when you can give a folder, scope, safety gates, and verification requirements.
Operating system.
Use for skills, memory, scheduled jobs, local files, tools, messaging, and repeatable business workflows.
Automation readiness wizard
Find the safest first automation.
Answer six beginner questions. The wizard outputs the right starting level, safest first task, model/tool choice, review gate, and copy/paste prompt. It never recommends sending, publishing, buying, account changes, or record changes as the first move.
Copy/paste prompt
One-job automation prompt.
I want to automate one safe business prep task. Ask me only for missing details that change the workflow. Then return: job name, approved inputs, output format, what AI may do, what AI must not do, human approval gate, verification checklist, and first 20-minute setup step.
Task: [describe the repeated task]
Safety: AI may draft, organize, summarize, and create local files. Human approval is required before sending, publishing, buying, account changes, record changes, or external actions.Copy/paste prompt
Codex / OpenClaw build brief.
You are working in this local project: [path]. First inspect project state. Build the smallest useful local artifact for this workflow. Do not deploy, publish, send outreach, buy, create accounts, accept terms, or use private credentials. Run tests or verification. Report exact files changed and exact command output.Download
Prompt pack + worksheets.
Includes one-job worksheet, skill builder prompts, agent job cards, model routing cards, Hermes cron prompt templates, and safety checklist.
Download ZIPOpen printable packOfficial Hermes path
Install, setup, then skill.
Official docs show the CLI install path and setup flow. Start with Hermes setup, enable only the tools you need, then save workflows as skills once they work.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup --portal
hermes doctor
hermes tools
hermes skills list
AI may
Prepare the work.
Draft, organize, summarize, score, route, remind, create local files, make checklists, and prepare review notes from approved inputs.
Human approval lock
External actions wait.
Sending, publishing, contacting, buying, accounts, terms, deployments, deleting, record changes, private credentials, legal/tax/payment/KYC/identity data, and compliance-sensitive claims require explicit human approval.
Next manual path
After this, build the actual business system.
Use Guide 03 for the smallest stack, Guide 04 for safe AI jobs, Guide 05 for content workflows, and Guide 06 for customer follow-up.
Guide 03Guide 04Guide library