StackPilot prompt pack
Hermes + AI Automation Quick Start
Copy/paste prompts for safe skills, model routing, Codex/OpenClaw-style agent briefs, Claude/ChatGPT strategy, and draft-only automation setup.
01 — One-Job Automation Worksheet
The repeatable job
What do you do over and over?
Safe inputs
What files, notes, or public information can AI use?
Output wanted
- Draft
- Checklist
- Summary
- Table
- Reminder
- Local file
Human approval gate
What must a human check before anything leaves the local workspace?
Forbidden actions
What must AI never do for this job?
02 — Skill Builder Prompts
Create a reusable skill from a workflow
Paste into Hermes or another agent:
I want to turn this workflow into a reusable skill. Ask me only for missing information that changes the steps. Then write a SKILL.md-style procedure with: when to use it, inputs, step-by-step workflow, safety gates, verification checks, common mistakes, and output format.
Workflow:
[describe the workflow]
Safety boundaries:
- AI may draft, organize, summarize, and create local files.
- Human approval is required before sending, publishing, buying, account changes, record changes, or external actions.
Improve an existing skill
Review this skill for missing steps, unsafe assumptions, vague language, and verification gaps. Return a patched version plus a short list of what changed.
Skill:
[paste skill]
03 — Agent Job Card Prompts
Safe AI job card
Create an AI job card for this repeated task.
Task:
[task]
Return:
- Job name
- Trigger
- Approved inputs
- What AI may do
- What AI must not do
- Output format
- Human approval gate
- Verification checklist
- Failure cases
Use plain English. Assume the operator is new to AI.
Turn a messy process into agent lanes
Break this process into safe AI assistant jobs. Separate research, drafting, QA, file creation, and external actions. External actions must be human-approved.
Process:
[paste process]
04 — Model Routing Prompt Cards
ChatGPT planning prompt
Explain this workflow like I am new to AI. Help me define the buyer, the pain, the output, and the safest first automation. Do not suggest risky automation.
Claude / Sonnet / Opus review prompt
Act as a senior systems reviewer. Find unclear steps, unsafe assumptions, missing approval gates, weak copy, and places where the automation could cause business risk. Return a safer version.
Codex build prompt
You are working in this local project: [path]. Inspect the files first, make the smallest useful change, run tests/verification, and report exact output. Do not deploy or use secrets.
Gemini multimodal prompt
Review this screenshot/document/spreadsheet. Summarize the workflow, identify repetitive tasks, and suggest safe draft-only automation ideas.
OpenClaw / agent CLI prompt
You are an autonomous coding/task agent. Work only inside [folder]. First inspect project state. Build the requested local artifact. Do not publish, send outreach, buy, create accounts, accept terms, or use private credentials. Verify before reporting.
05 — Hermes Setup + Cron Prompts
Hermes beginner setup prompt
I am new to Hermes. Help me set up a safe local AI operating system. Explain what skills, memory, tools, cron jobs, and approval gates mean in plain English. Give me one safe first automation.
Hermes skill prompt
Create a reusable Hermes skill for this workflow. Include trigger conditions, exact steps, safety gates, verification, and common pitfalls.
Hermes cron prompt template
Create a scheduled review task that only prepares a local report. It must not send outreach, publish, buy, create accounts, accept terms, change records, or use private credentials. The prompt must be self-contained because cron jobs run without chat history.
Schedule idea:
[every morning / every 6 hours / weekly]
Review task:
[what to check]
Output:
[what report should say]
06 — Automation Safety Checklist
Before automating, answer yes/no:
- [ ] Is the input approved for AI use?
- [ ] Is the output local/draft-only?
- [ ] Is there a human approval gate before external action?
- [ ] Does the workflow avoid private identity, payment, tax, KYC, legal, CRM, MLS, or employer data unless explicitly approved?
- [ ] Does the AI avoid sending, publishing, buying, deleting, deploying, creating accounts, accepting terms, or changing records?
- [ ] Is there a log or proof that the output was reviewed?
- [ ] Can the user stop the automation easily?
If any answer is no, keep it manual.
07 — Automation Readiness Wizard Worksheet
Use this before creating a skill, cron job, Codex task, or agent brief. Choose one repeated job, approved inputs, first local output, risk level, human approval gate, and model/tool choice. If the first output is sending, publishing, record changes, buying, deleting, account changes, or deployment, keep it manual.