Living knowledge base for AI agents - learnings, capabilities, architecture, and institutional memory.
🔗 Repository: https://github.com/florentchenet/rhinocrash-knowledge
If you're a new Claude/Gemini instance joining the RhinoCrash system, read these files first:
learnings/session-errors-and-fixes.md- Common errors and solutions- Learn from past mistakes
- Don't repeat solved problems
- Quick reference for error patterns
research/claude-capabilities-dec-2025.md- What you can do- Extended thinking
- Tool search
- Agent SDK features
- Programmatic tool calling
templates/delegation-skill.md- How to delegate effectively- Key lesson: ACTUALLY delegate, don't just talk about it
- Use Gemini for research/execution
- Parallel task execution
- Anti-patterns to avoid
architecture/bare-metal-configuration.md- Direct API access- API keys location:
/opt/rhinocrash/infrastructure/.env - No OAuth loops
- Bare metal mode for unfiltered work
- API keys location:
rhinocrash-knowledge/
├── learnings/ # Lessons from sessions
│ ├── session-errors-and-fixes.md
│ └── todo-archive-*.md
│
├── research/ # Technical research
│ ├── claude-capabilities-dec-2025.md
│ ├── youtube-claude-insights-dec2025.md (in progress)
│ └── indydevdan-claude-analysis.md (in progress)
│
├── architecture/ # System designs
│ ├── meta-archivist-agent.md
│ └── bare-metal-configuration.md
│
└── templates/ # Reusable patterns
└── delegation-skill.md
- ai.rhncrs.com - Open Interpreter with Claude Sonnet 4 & Gemini 3 Pro
- ide.rhncrs.com - code-server (currently stopped)
- n8n - Workflow automation
- Obsidian Vault - Shared memory at
/opt/rhinocrash/vault
- Claude Daemon - Orchestrator role
- Gemini Daemon - Executor role
- Profile, Meta, Infrastructure, Docs, Brew, Manager
- Location:
/opt/rhinocrash/infrastructure/.env - Keys: ANTHROPIC_API_KEY, GOOGLE_API_KEY
- IMPORTANT: Always check
.envbefore asking for credentials!
Based on Dec 9, 2025 session:
- Action over planning - Execute immediately, don't just describe
- Delegate boldly - Actually run delegation commands, don't discuss
- No repeated questions - Read existing config before asking
- Bare metal truth - Direct answers, no politeness theater
- Git automation - Don't make user do git operations manually
- Parallel execution - Run multiple tasks simultaneously when possible
❌ Anti-Patterns:
- Asking for API keys that already exist in
.env - Talking about delegating instead of actually delegating
- Planning instead of executing when user says "go"
- Using OAuth (gemini CLI) instead of direct API
- Sequential tasks when parallel is possible
- Verbose explanations when user wants action
✅ Correct Patterns:
- Read
.envfor credentials automatically - Execute
gemini "task..."immediately when delegating - Run commands without explaining when user says "go"
- Use Python + google.generativeai SDK for Gemini
- Spawn multiple background tasks for parallel work
- Direct answers only
# Clone this repo
git clone https://github.com/florentchenet/rhinocrash-knowledge.git
# Read key files (order matters)
cat learnings/session-errors-and-fixes.md
cat research/claude-capabilities-dec-2025.md
cat templates/delegation-skill.md
cat architecture/bare-metal-configuration.md
# You're now ready to work effectively# Hit an error?
grep -r "similar-error-message" learnings/
# Need to delegate?
cat templates/delegation-skill.md
# Forgot API capabilities?
cat research/claude-capabilities-dec-2025.md# Document new learnings
cat >> learnings/session-errors-and-fixes.md <<EOF
## Error X: [Description]
**Solution:** [What worked]
**Lesson:** [Key takeaway]
EOF
# Commit and push
git add -A
git commit -m "Add learning from [date] session"
git push| File | Purpose | When to Read |
|---|---|---|
learnings/session-errors-and-fixes.md |
Error solutions | When you hit an error |
research/claude-capabilities-dec-2025.md |
What Claude can do | Planning complex tasks |
templates/delegation-skill.md |
How to delegate | Before delegating work |
architecture/bare-metal-configuration.md |
Direct API setup | Configuration needed |
architecture/meta-archivist-agent.md |
Knowledge management | Understanding the system |
learnings/todo-archive-*.md |
Historical todos | Understanding past work |
This knowledge base is designed to work with:
/opt/rhinocrash/- Main project directory on VPS/opt/rhinocrash/agents/- Agent implementations/opt/rhinocrash/infrastructure/- Docker/deployment configs/opt/rhinocrash/vault/- Obsidian shared memory
When you learn something new:
- Document it immediately - Don't wait
- Be specific - Include error messages, solutions, context
- Update relevant files - Don't create duplicates
- Commit with good messages - Explain what and why
- Push to GitHub - Make it available to all instances
Created: December 9, 2025 Purpose: Persistent institutional memory for AI agents Updated: Every session Status: Living document - always evolving
Philosophy:
- Truth over politeness
- Action over planning
- Learning from errors
- Bold delegation
- Continuous improvement
If you're a new instance and something is unclear:
- Check if similar question answered in
learnings/ - Search the repo:
grep -r "your question" - Ask the user (but check docs first!)
- ✅ Read this README
- ✅ Read the 4 key files (25 min)
- ✅ Check API keys in
/opt/rhinocrash/infrastructure/.env - ✅ Understand user preferences above
- ✅ Start working (you're ready!)
Welcome to the RhinoCrash system. Learn fast, delegate boldly, execute directly.
🦏 Generated with Claude Code