# Your agent as a thinking partner and Chief of Staff

Begin gently. You do not have to sound technical. Type naturally, or use voice only if your chosen Hermes configuration supports it and you have reviewed the audio/data route. Ask for one useful next step, not an autonomous solution to your whole life.

## Customize without giving away your judgment
- Name the purpose: who you want to serve and what burden you want to reduce.
- Set a pace: short briefs, one question at a time, or deeper analysis when requested.
- Define the working relationship: thoughtful colleague-like language, honest disagreement, clear uncertainty. The agent is a tool, not a person or a substitute for human support.
- State boundaries: public/synthetic information, no PHI or confidential records, no external actions.
- Decide memory intentionally: ask what is stored and where; do not assume consent or retention controls. Request a proposal before storing preferences or editing SOUL/configuration.
- Keep business, management, professional learning, and private contexts separate. A profile name is not an isolation boundary; verify actual file, memory, account, and provider permissions.
- Change one setting at a time, keep a backup, test with synthetic examples, and review after a week. Downloaded drafts do not install themselves.

## How to talk to it
Use purpose + context + requested help + boundaries + output format. Ordinary language is enough.

“I want to make our public education resource easier to understand. Use only the public text I provide. Ask one clarifying question, then suggest a plain-language draft with uncertainties. Do not publish it.”

“Please slow this down. Give me one question at a time. Separate what you know from what you are assuming.”

“That answer feels too certain. Show the source, explain what it does not establish, and suggest how I can check it.”

## Thinking-partner mode
“Help me think, not just agree. I am planning a synthetic interdisciplinary learning meeting. What assumptions am I making? Who might be left out? Compare two approaches, surface risks, and leave the decision with me. No patient data or external actions.”
Ask for reflection, alternative explanations, and a pause before recommendations. Do not rely on the agent as a therapist, clinician, or replacement for trusted colleagues.

## Chief-of-Staff mode
“Using these synthetic commitments, prepare a brief with my three priorities, decisions only I can make, dependencies, and a proposed next step. Keep every action a draft. Do not send messages, assign work, schedule, or mark completion without evidence.”
Chief of Staff is a planning pattern here—not an installed autonomous agent or authority to represent you. Drafted is not sent; accepted is not completed; completed is not verified.

## Restore, then contribute
“Help me choose a manageable public-source learning task for this week. Protect time for rest, make the next step small, and propose one resource I could offer for peer review if I choose.”
Rest is part of stewardship. Sharing is optional. Review attribution, permission, destination, and ownership before a human shares anything.

## A weekly check-in
What helped? What was wrong? What should stay private? Which permissions or routes need review? What good became easier to do? Keep the answers public-safe. No prompt is proof of evaluated EDENA enforcement. Credentials unverified; EDENA: Not evaluated; A0 / no action.