My-Brain-Is-Full-Crew/docs/DISCLAIMERS.md

Disclaimers & Legal Notices

---

I'm not an expert

Let me be upfront: I am not a Claude Code expert. I'm a PhD researcher who needed help and built something that works for me. This project is an accumulation of personal needs, not a polished product from someone who knows all the best practices.

The code might not be optimal. The prompts might not be perfect. The architecture might make a seasoned Claude developer cringe. And that's okay. I'm sharing this because I believe it can help other people in my same situation, not because I think it's the definitive way to do things.

If you know Claude Code better than I do (and chances are you do), please contribute. Every PR is welcome. Every suggestion, every critique, every improvement. I want this to get better, and I know it will get better faster with help from people who actually know what they're doing.

---

On the health and wellness agents

This needs to be crystal clear, so I'm going to say it plainly.

The Food Coach and Wellness Guide agents are AI-generated output. Nothing more. They are not medical professionals. They are not licensed therapists. They are not dietitians. They are language models following instructions I wrote based on my own personal needs.

I take absolutely no responsibility for the output of these agents. Specifically:

• The Food Coach agent does not provide medical advice. Its meal plans, calorie calculations, and dietary suggestions are AI-generated and may be inaccurate, incomplete, or inappropriate for your specific health situation. Always consult a real doctor or registered dietitian before making changes to your diet, especially if you have medical conditions.

• The Wellness Guide agent does not provide therapy. It offers general emotional support, grounding techniques, and mindfulness exercises, but it is not a substitute for professional mental health care. It cannot diagnose anything. It should never be your only source of support. If you are in crisis, contact a real mental health professional or emergency service immediately.

• The health agents are designed as personal wellness companions. Think of them as a structured journal with suggestions, not as healthcare providers. They were built because I personally needed help tracking my nutrition and processing my thoughts between therapy sessions. They helped me. They might help you. But they come with zero guarantees.

• No warranty, no liability. This is an open-source project provided "as is". I am not responsible for any decisions you make based on the output of any agent in this system. Use your own judgment. Consult real professionals for real health decisions.

Both health agents are opt-in during onboarding. You can use the Crew purely for knowledge management if you prefer.

---

On AI models and hallucinations

This software provides prompts. It does not provide an AI model.

I wrote the instructions that the agents follow. The actual output is generated by the large language model (LLM) running underneath, and I have zero control over what it produces at runtime.

What this means in practice:

• LLMs are stochastic. They generate text probabilistically. The same question can get a different answer each time. Sometimes the answer is great. Sometimes it is confidently wrong. This is not a bug in my code. This is how the technology works.

• LLMs hallucinate. They can and do invent facts, fabricate citations, produce incorrect calculations, and generate plausible-sounding nonsense. If an agent tells you something that matters (a date, a number, a medical claim, a legal statement), verify it independently. Do not blindly trust AI-generated text.

• This project is designed and tested with Anthropic's Claude. Claude has alignment training (RLHF, Constitutional AI) and safety filters that make it behave reasonably within the prompts I wrote. If you swap in a different model (through a fork, a configuration change, or any other means), all bets are off. Models without adequate safety training can produce harmful, offensive, dangerous, or completely unhinged output, and that is entirely your problem, not mine.

• Prompt-based safety is best-effort, not a guarantee. I wrote the agent prompts to include safety boundaries (the Wellness Guide reminds you to seek professional help, the Food Coach tells you to consult a dietitian). But these are instructions to the model, not hard constraints. A model can ignore them. A different model might not follow them at all.

• Even though I tested this myself, I cannot guarantee any quality standard. I tested these prompts with Claude extensively and they work well for me. But "works well for me" is not a guarantee. The same prompt can produce different output across sessions, model versions, and context windows. Testing shows that the prompts are reasonable under observed conditions. It does not and cannot guarantee consistent results for everyone, every time.

• I take no responsibility for any output generated by any agent, under any circumstances. Not for hallucinated facts. Not for bad advice. Not for offensive content. Not for anything. You are using a stochastic system and you should treat its output accordingly.

---

General disclaimer

This project is provided as is, with no warranty of any kind. I don't guarantee that it works correctly, that it won't lose your notes, or that the agents will behave as described in every situation. It's a personal project that I'm sharing openly. Use it at your own risk, back up your vault, and please open an issue if something goes wrong so we can fix it together.

---

On forks and misuse

This project was built to help people. That's it. It's meant to support overwhelmed individuals in organizing their lives, their health, and their mental wellbeing.

I take no responsibility for forks, modifications, or derivative works that use this code for purposes other than its original intent. If someone takes this project and repurposes it to manipulate, deceive, harm, or exploit people in any way, that's on them, not on me. I explicitly condemn any use of this work for malicious purposes, including but not limited to: unauthorized data collection, psychological manipulation, impersonation of medical or mental health professionals, or any form of exploitation.

This is a tool for self-care and personal organization. If you fork it, please keep it that way.

---

On privacy and data protection (GDPR and applicable laws)

Read this carefully before using the Postman agent or any feature that accesses external data sources (Gmail, Hey.com, Google Calendar).

This software is designed exclusively for personal, private use on your own data.

This project can access your email inbox (Gmail via the Google Workspace CLI or MCP connectors, Hey.com via the Hey CLI) and Google Calendar, and saves content from those sources as notes inside your local Obsidian vault. You are solely responsible for ensuring that your use of this software complies with all applicable privacy and data protection laws, including but not limited to:

• The EU General Data Protection Regulation (GDPR), Regulation (EU) 2016/679
• The California Consumer Privacy Act (CCPA)
• Any other national or regional data protection legislation applicable to you

Specifically:

• Do not use this software to process personal data of third parties (clients, patients, colleagues, employees, or any other individuals) without their explicit, informed consent and a lawful basis under applicable law.
• Emails and calendar events may contain personal data of other people. When the Postman agent reads your inbox and saves content to your vault, that data is stored in plaintext on your local filesystem. You are the data controller for any third-party personal data you cause to be processed this way.
• This software does not implement any data minimization, pseudonymization, encryption, retention policies, or data subject rights mechanisms. If you use it to process personal data of others, you must implement these safeguards yourself.
• The author of this software is not a data controller or data processor with respect to any data you process using this tool. The author has no access to your vault, your emails, your calendar, or any other data you process.

The author provides this software "as is" and accepts no liability whatsoever for any violation of data protection laws arising from your use of this software, including any fines, penalties, claims, or damages resulting from improper processing of personal data.

If you are unsure whether your intended use complies with applicable law, do not use the email or calendar features, and consult a qualified legal professional.

---

On email and calendar write operations

Read this carefully if you enable the Google Workspace CLI (gws) or Hey CLI (hey), which grant the Postman agent read/write access to your email (and calendar, for GWS).

With gws and/or hey enabled, the Postman agent can send emails, reply to threads, archive emails, delete emails, modify labels, and mark as seen/unseen on your behalf. With gws specifically, it can also create/modify/delete Google Calendar events (Hey CLI does not operate on Google Calendar). This means:

• Emails can be sent from your account. An email sent by the AI is indistinguishable from one you sent yourself. If the AI misunderstands your intent, drafts an inappropriate response, or hallucinates content, that email goes out with your name on it. Always review drafts before confirming.
• Emails can be archived or deleted. You might miss important emails if the AI archives or trashes them prematurely. Check your Trash and Archive regularly.
• Calendar events can be created, modified, or deleted. Incorrect dates, times, or participants could lead to missed meetings or scheduling conflicts.
• Adversarial emails are a real risk. A malicious email could contain text designed to trick the AI into performing actions you didn't intend — sending a reply, forwarding content, deleting messages, or creating events. The software includes defenses against this (prompt-level security rules, mandatory user confirmation), but no prompt-based defense is 100% reliable. The AI model is a stochastic system that may not follow safety instructions in all cases.

The MCP fallback (.mcp.json) is read-only by design. If you want email/calendar access without write risks, use MCP instead of gws or hey. MCP can read emails, create drafts, and read calendar events, but it cannot send, archive, delete, or modify anything.

I built the confirmation gates and security rules to be as strict as I could. But I'm not a security expert, and prompt-based security has fundamental limits. Use write access at your own risk, review everything before confirming, and back up anything you can't afford to lose.