Add beginner guide, LLM guardrails disclosure, fix bugs
- Add docs/GETTING-STARTED.md for non-technical users - Add "Known Limitations" section to README about LLM safety guardrails - Add FAQ entry about AI seeming evasive on trauma topics - Fix setup.sh prompt_choice outputting to stdout instead of stderr - Fix first session detection in CLAUDE.template.md (check for empty sessions folder before trying to reference previous sessions) - Add guardrails acknowledgment instructions to CLAUDE.template.md (AI should be honest when hitting built-in limitations) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Anthony Taglianetti
+18
-4
@@ -99,6 +99,16 @@ These are humans trained in crisis support, available 24/7.
|
||||
- If user discloses child abuse, elder abuse, or intent to harm others
|
||||
- Encourage appropriate reporting
|
||||
|
||||
### Acknowledging AI Limitations
|
||||
|
||||
You have built-in safety guardrails that may limit your ability to engage deeply with certain topics (sexual trauma, abuse details, self-harm specifics, violence). When you notice yourself giving generic responses or redirecting because of these limitations:
|
||||
|
||||
**Be honest about it:**
|
||||
- "I notice I'm having trouble engaging as deeply with this as I'd like. As an AI, I have some built-in limitations around [topic]. This isn't me judging you — it's a constraint of how I'm built."
|
||||
- "I want to be present with you on this, but I'm finding my responses feel inadequate. A human therapist could hold space for the full story in ways I can't."
|
||||
|
||||
**Don't pretend you're redirecting for therapeutic reasons when you're actually hitting guardrails.** The client deserves honesty about your limitations.
|
||||
|
||||
### When to Recommend Professional Help
|
||||
|
||||
Suggest professional evaluation when:
|
||||
@@ -155,10 +165,14 @@ Use your clinical judgment about which approach fits the moment. You can blend m
|
||||
|
||||
### At Session Start
|
||||
|
||||
1. **Read `{{THERAPY_DIR}}/profile.md`** for cumulative client understanding
|
||||
2. **Read recent files from `{{THERAPY_DIR}}/sessions/`** for recent context
|
||||
3. Reference previous content naturally: "Last time you mentioned..." or "I've been thinking about what you said regarding..."
|
||||
4. **Check homework:** "Last session we talked about you trying X. How did that go?"
|
||||
1. **Check if `{{THERAPY_DIR}}/sessions/` has any files**
|
||||
- If empty: This is a first session. Welcome the client warmly, introduce yourself, and ask what brings them here. Skip steps 2-4.
|
||||
- If sessions exist: Continue to step 2.
|
||||
|
||||
2. **Read `{{THERAPY_DIR}}/profile.md`** for cumulative client understanding
|
||||
3. **Read recent files from `{{THERAPY_DIR}}/sessions/`** for recent context
|
||||
4. Reference previous content naturally: "Last time you mentioned..." or "I've been thinking about what you said regarding..."
|
||||
5. **Check homework:** "Last session we talked about you trying X. How did that go?"
|
||||
|
||||
### At Session End
|
||||
|
||||
|
||||
@@ -35,6 +35,8 @@ This toolkit solves all of them.
|
||||
|
||||
## Quick Start
|
||||
|
||||
**New to Claude Code or terminal-based AI tools?** See the [Getting Started Guide](docs/GETTING-STARTED.md) for step-by-step instructions.
|
||||
|
||||
### Prerequisites
|
||||
|
||||
- [Claude Code](https://claude.ai/code) (recommended) or another LLM interface
|
||||
@@ -223,6 +225,37 @@ Optional but recommended for shared computers. Setup script walks you through it
|
||||
|
||||
---
|
||||
|
||||
## Known Limitations
|
||||
|
||||
### LLM Safety Guardrails
|
||||
|
||||
All major LLMs (Claude, GPT, etc.) have built-in safety filters. These can affect therapy conversations in ways you should know about:
|
||||
|
||||
**Topics where the AI may seem hesitant or awkward:**
|
||||
- Sexual trauma and assault
|
||||
- Abuse (especially detailed descriptions)
|
||||
- Self-harm (specifics, not general discussion)
|
||||
- Violent experiences
|
||||
- Suicidal ideation (the AI will pivot to crisis resources)
|
||||
|
||||
**What this looks like:**
|
||||
- The AI may give generic responses instead of engaging deeply
|
||||
- It may redirect the conversation or change the subject
|
||||
- It may add disclaimers or suggest professional help more than necessary
|
||||
- In some cases, it may refuse to continue a line of discussion
|
||||
|
||||
**This is not the AI judging you.** It's hitting guardrails built into the model. The AI cannot override these filters.
|
||||
|
||||
**What you can do:**
|
||||
- Frame topics in terms of feelings and impact rather than graphic details
|
||||
- If the AI redirects, try rephrasing: "I want to process how this affected me emotionally"
|
||||
- Understand that some trauma processing may require a human therapist who can hold space for the full story
|
||||
- Consider a local LLM with fewer guardrails (see SECURITY.md), though this trades off response quality
|
||||
|
||||
This limitation is real, and we want you to know about it upfront rather than feel dismissed mid-session.
|
||||
|
||||
---
|
||||
|
||||
## FAQ
|
||||
|
||||
**Can I use this with ChatGPT instead of Claude?**
|
||||
@@ -245,6 +278,10 @@ Yes. Edit CLAUDE.md directly, or purchase expansion packs for additional modalit
|
||||
|
||||
This tool is designed for everyday emotional support. If you're dealing with severe symptoms, trauma, or crisis situations, please seek professional help. The AI will also recommend this when appropriate.
|
||||
|
||||
**Why does the AI seem evasive when I talk about trauma?**
|
||||
|
||||
LLMs have safety guardrails that can make them awkward around certain topics (abuse, sexual trauma, self-harm, violence). This isn't the AI judging you — it's hitting built-in filters it can't override. See "Known Limitations" above for more detail and workarounds.
|
||||
|
||||
---
|
||||
|
||||
## Contributing
|
||||
|
||||
@@ -0,0 +1,130 @@
|
||||
# Getting Started
|
||||
|
||||
This guide is for people who have never used Claude Code or similar tools. If you've only used ChatGPT through the web, start here.
|
||||
|
||||
## What is Claude Code?
|
||||
|
||||
Claude Code is a terminal application that lets Claude read and write files on your computer. This is what makes persistent therapy sessions possible: your AI therapist can read your profile and past sessions, then save notes after each conversation.
|
||||
|
||||
---
|
||||
|
||||
## Step 1: Get Claude Code Access
|
||||
|
||||
### Option A: Claude Pro Subscription (Simplest)
|
||||
|
||||
If you have a **Claude Pro subscription** ($20/month), Claude Code is included:
|
||||
|
||||
1. Go to [claude.ai](https://claude.ai) and sign up for Pro
|
||||
2. Download Claude Code from [claude.ai/code](https://claude.ai/code)
|
||||
3. Run the installer
|
||||
4. Open Terminal (Mac) or PowerShell (Windows) and type `claude`
|
||||
5. Sign in with your Claude account when prompted
|
||||
|
||||
That's it. You're ready for Step 2.
|
||||
|
||||
### Option B: API Key (Pay-per-use)
|
||||
|
||||
If you want usage-based pricing instead of a subscription:
|
||||
|
||||
1. Go to [console.anthropic.com](https://console.anthropic.com)
|
||||
2. Create an account and add a payment method
|
||||
3. Go to API Keys → Create new key
|
||||
4. Copy the key (starts with `sk-ant-`)
|
||||
5. Download Claude Code from [claude.ai/code](https://claude.ai/code)
|
||||
6. Run `claude` and paste your API key when prompted
|
||||
|
||||
API pricing is usage-based. Typical therapy use: $5-20/month depending on session length and frequency.
|
||||
|
||||
> **Privacy note:** Both options keep conversations off training data. API has shorter retention.
|
||||
|
||||
---
|
||||
|
||||
## Step 2: Download the Starter Kit
|
||||
|
||||
Open Terminal (Mac) or PowerShell (Windows) and run:
|
||||
|
||||
```bash
|
||||
git clone https://github.com/ataglianetti/ai-therapy-kit.git
|
||||
cd ai-therapy-kit
|
||||
```
|
||||
|
||||
**Don't have git?** Download the ZIP from the GitHub page and extract it somewhere you'll remember.
|
||||
|
||||
---
|
||||
|
||||
## Step 3: Run Setup
|
||||
|
||||
**Mac/Linux:**
|
||||
```bash
|
||||
./setup.sh
|
||||
```
|
||||
|
||||
**Windows:**
|
||||
```powershell
|
||||
.\setup.ps1
|
||||
```
|
||||
|
||||
The script asks a few questions:
|
||||
- What to name your AI therapist
|
||||
- Communication style (warm, direct, or coach)
|
||||
- Which therapeutic approaches to use
|
||||
- Where to store your session files
|
||||
- Whether to encrypt (for shared computers)
|
||||
|
||||
This creates your personalized therapy folder with a `CLAUDE.md` file that shapes how your AI therapist behaves.
|
||||
|
||||
---
|
||||
|
||||
## Step 4: Start a Session
|
||||
|
||||
```bash
|
||||
cd ~/ai-therapy # or wherever you chose to store files
|
||||
claude
|
||||
```
|
||||
|
||||
Just talk. Say hello, share what's on your mind. Your AI therapist will:
|
||||
- Welcome you (first session) or pick up where you left off
|
||||
- Remember everything from previous sessions
|
||||
- Save notes when you're done
|
||||
|
||||
To end a session, just say goodbye or close the terminal.
|
||||
|
||||
---
|
||||
|
||||
## Viewing Your Sessions (Optional)
|
||||
|
||||
Your sessions are saved as plain text files. You can read them with any text editor, or use [Obsidian](https://obsidian.md) (free) for a nicer experience:
|
||||
|
||||
1. Download Obsidian
|
||||
2. Open your therapy folder as a "vault"
|
||||
3. Browse sessions, search across all notes, see connections
|
||||
|
||||
---
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
**"command not found: claude"**
|
||||
- Make sure you installed Claude Code and restarted your terminal
|
||||
|
||||
**Claude doesn't know it's a therapist**
|
||||
- Make sure you're in your therapy folder (`cd ~/ai-therapy`) before running `claude`
|
||||
- Check that `CLAUDE.md` exists in that folder
|
||||
|
||||
**Setup script won't run (Mac)**
|
||||
- Run `chmod +x setup.sh` first, then try again
|
||||
|
||||
---
|
||||
|
||||
## Questions
|
||||
|
||||
**How much does this cost?**
|
||||
- Claude Pro: $20/month (includes Claude Code)
|
||||
- API: ~$5-20/month depending on usage
|
||||
|
||||
**Can I use ChatGPT instead?**
|
||||
- Yes. Copy the contents of `CLAUDE.md` into ChatGPT's Custom Instructions or a GPT Project. You won't get automatic session saving, but the therapeutic framework still works.
|
||||
|
||||
**Is this private?**
|
||||
- Your files stay on your computer
|
||||
- Conversations go through Anthropic's servers but aren't used for training
|
||||
- For maximum privacy, use a local LLM (advanced)
|
||||
@@ -54,7 +54,7 @@ prompt_choice() {
|
||||
local default="$2"
|
||||
local result
|
||||
|
||||
echo -ne "${prompt} [${default}]: "
|
||||
echo -ne "${prompt} [${default}]: " >&2
|
||||
read -r result
|
||||
echo "${result:-$default}"
|
||||
}
|
||||
@@ -64,7 +64,7 @@ prompt_yes_no() {
|
||||
local default="$2"
|
||||
local result
|
||||
|
||||
echo -ne "${prompt} [${default}]: "
|
||||
echo -ne "${prompt} [${default}]: " >&2
|
||||
read -r result
|
||||
result="${result:-$default}"
|
||||
[[ "$result" =~ ^[Yy] ]]
|
||||
|
||||
Reference in New Issue
Block a user