Agent Playbooks

Which book is for you?

Two books, one system. Simple test: have you installed Hermes yet?

No → start with *Hermes Agent for Complete Beginners* ($12). Zero experience assumed: install in 15 minutes, first useful conversations in a weekend, plain-English glossary. It deliberately doesn't teach tuning — it gets you to *using it daily*.

Yes, and you're hitting edges — config sprawl, cost surprises, memory that lies → *58 Prompts for Power Users* ($29). Paste-ready fixes, every one citing the incident that produced it. Not beginner material, on purpose.

Want both? The Complete Collection ($35) has them together, plus every future update.

Honest note, taken from the books themselves: if you installed Hermes last weekend, Book 1 won't help you yet — Appendix D says so in Book 2's own words. Give it a month. The prompts keep.

What's in the full 58

Hermes Agent: 58 Prompts for Power Users — $29

The 58 prompts behind The Hermes Guide, in paste-ready form:

• 14 categories mapped to the job you're doing, not the API: Machine Setup, Cost & Speed, Context, Memory, Skills, Security, Safety, Proactivity, Working Style, Personality, Multi-Agent, Gateway, Troubleshooting, General

• Every prompt cites its source — the thread, the user, the incident that produced it

• The ★ 16 flagged — if you only ever run 16 things, run these

• Searchable library that works offline — filter by category, star, or keyword

• Free updates — when prompts get better, you get the new version

The Prompt Library here is free because audits and self-review prompts should be free. The other 42 are the ones people paid for the book to get — the cost forensics, the gateway lockdown, the multi-agent verification chains.

One-time purchase. No subscription, no seat limits, no upsell inside.

Prompt of the Week #1: The hardware & environment audit

Every week: one prompt in full, what it does, what to expect.

This week — the one to run first:

```

Audit this machine and save what matters to your memory.

Run real commands to find out: OS and version, CPU/architecture, total RAM,

free disk space, GPU/VRAM if any, Python version and whether PEP 668 applies,

Node version, package managers available (brew/apt/winget), whether Docker is

installed and running, and whether Ollama or LM Studio is present with which

models pulled.

Then save ONLY the durable facts to memory as declarative statements — things

that will still be true next month and that change how you should work here.

Skip anything easily re-discovered in one command. Tell me exactly what you

saved and why each entry earns its space.

```

Why it's first: every other prompt in the set assumes the agent knows the machine. Most bad agent behavior — wrong package manager, wrong assumptions about RAM, "fixing" things that aren't broken — traces back to skipping this step.

What to expect: the agent runs 8–12 real commands, then writes 5–10 memory lines. Check them. Anything that won't be true next month doesn't belong — delete it and tell the agent why.

Post your output below — especially the memory lines you deleted. Those are the interesting ones.

Read this before you paste

These prompts are powerful on purpose — several of them instruct your agent to change its own configuration, kill processes, or rewrite memory. Three rules:

1. Read the whole prompt before you send it. You should know what you're asking for. Nothing here does anything the prompt doesn't say out loud.

2. The audit prompts are safe-by-default; the fix prompts are not automatic. Audits report. Fixes (like "Gateway lockdown" or "The isolation plan") change things — run them when you understand the change they'll make.

3. Never let an agent be its own witness. That's why "Verify the subagent, don't trust it" and "Is it actually working, or just saying so?" exist. If your agent claims something happened, require the artifact — the file, the log line, the URL.

Broken something? Post the failure here. A prompt that failed honestly is worth ten that worked.