by lout33
A symbiotic AI agent that remembers everything, challenges you, and extends your cognition.
# Add to your Claude Code skills
git clone https://github.com/lout33/claude_life_assistantA symbiotic AI agent that remembers everything, challenges you, and extends your cognition.
Two files that turn Claude into a symbiotic agent:
The key difference: Not a chatbot. An agent that lives in your filesystem, remembers your context, sees patterns you miss, and operates alongside you.
graph LR
A[You] <-->|Symbiotic| B[Agent]
B -->|Reads| C[CLAUDE.md<br/>Who you are]
B -->|Reads| D[NOW.md<br/>Current state]
B -->|Updates| E[Memory Log<br/>Patterns over time]
B -->|Acts| F[Code/Files/Research]
E -.->|Part of| D
style C fill:#2d3748,stroke:#4a5568,color:#fff
style D fill:#2d3748,stroke:#4a5568,color:#fff
style E fill:#2d3748,stroke:#4a5568,color:#fff
style A fill:#2d3748,stroke:#4a5568,color:#fff
style B fill:#2d3748,stroke:#4a5568,color:#fff
style F fill:#2d3748,stroke:#4a5568,color:#fff
The Philosophy: Symbiotic AI through transparency. All memory lives in files you control. The agent builds context over time, challenges your blind spots, and you see everything it remembers.
One command:
curl -fsSL https://raw.githubusercontent.com/lout33/claude_life_assistant/main/install.sh | bash
Works with Claude Code, OpenCode, or as a local project. The script auto-detects your setup.
No comments yet. Be the first to share your thoughts!
Windows users: Run in WSL or Git Bash.
Option 1: Clone
git clone https://github.com/lout33/claude_life_assistant
cd claude_life_assistant
Then open with Claude Code and start talking.
Option 2: Global Install (Claude Code)
cp CLAUDE.md NOW.md ~/.claude/
Option 3: Global Install (OpenCode)
cp CLAUDE.md ~/.config/opencode/AGENTS.md
cp NOW.md ~/.config/opencode/
Note: Rename CLAUDE.md to AGENTS.md for OpenCode.
After install, just start talking. That's it.
The agent reads your files at session start. It knows your identity, your current projects, your patterns. No commands needed — it adapts to whatever you're doing.
Built-in commands for daily rituals:
| Command | What it does |
|---------|--------------|
| /start-day | Morning kickoff. Sets your MIT (Most Important Task) for the day. |
| /check-day | Quick check-in anytime. Accountability without friction. |
| /end-day | Evening review. Captures wins, lessons, updates the log. |
| /reflect | Deep reflection. Creates journal entries, surfaces patterns. |
Just type the command. The agent handles the rest.
| File | Purpose |
|------|---------|
| CLAUDE.md / AGENTS.md | Who you are, how you work, your known bugs |
| NOW.md | Current mode, this week's actions, Memory Log |
Note: Claude Code uses CLAUDE.md, OpenCode uses AGENTS.md. The install script handles this automatically.
Why 2 files?
Separation = clarity. CLAUDE.md is your operating system (identity, psychology, patterns). NOW.md is your current state (projects, tasks, memory). The agent reads both, updates NOW.md as you work.
Conversation > Documentation
You talk. The agent acts and maintains the files. No manual editing required (though you can if you want).
Most AI validates. This one calls you out.
From a real conversation:
AI: "You find something valuable → People want it → You feel repulsed by the exchange → You give it away for free → You have no money → Repeat."
AI: "That's not idealism. That's self-punishment."
The instruction is simple: "No coddling. Quote his words back when off track."
This isn't a one-off ChatGPT session. The agent has persistent memory — your values, your goals, your known bugs, your history.
When it says "You did this with the Blender addon — paying customers, real traction, and you gave it away," it's not guessing. It remembered.
Not just chat. The agent executes: writes code, researches, creates files, runs commands. It operates alongside you, not just advises.
Local:
CLAUDE.md, NOW.md) are stored on your machine.gitignore for personal files)Cloud:
Want full privacy? This file-based system works with any LLM. For local-only processing, use with Ollama or LM Studio.
In NOW.md, the agent maintains a dated log:
This is what makes it symbiotic. The agent sees patterns you miss, remembers what you forget, and builds context across sessions.
Example progression:
### Week 1, Jan 7
- Setup complete. Known bug: 'I keep adding features instead of launching'
### Week 1, Jan 10
- Avoided user call. Rescheduled twice. Pattern: building = safe, talking = scary.
### Week 2, Jan 15
- Had first user call. Quote: 'I've been building what I think they want instead of asking'
### Week 4, Feb 1
- Pattern confirmed: 3 weeks of work on feature nobody asked for. This is the 3rd time.
The longer you use it, the better it gets.
Check examples/ for real progressions:
Each shows Week 1 to Month 3 evolution with Memory Log progression.
You want AI that knows you and challenges you:
You probably won't like this if:
Symbiotic > Assistive — The agent operates with you, not for you
Challenge > Validate — Honest reflection beats comfortable agreement
Memory compounds — The longer you use it, the better it gets
Ship ugly — Done beats perfect
This template covers the basics. My personal setup includes:
Interested in the full system? Let me know (feedback helps prioritize what to build next).
Part of the symbiotic AI ecosystem:
| Project | What it does | |---------|--------------| | what-did-i-do | Know what you actually did today. Passive screen tracker that captures screenshots, analyzes with Gemini Vision, and creates daily activity reports. | | writing-style-skill | Make AI write like you. Export your emails/chats, extract your patterns, transform any AI text into your voice. |
More skills coming. Star to follow.
Created by @lout33
Built on the belief that symbiotic AI means shared cognition, not supervision. The agent operates with you, not for you.
"The solution isn't a better system. It's doing the uncomfortable thing you're avoiding."
