roboco/docs/rag/workflows/cross-agent-learning.md

Cross-Agent Learning

When one agent learns something, all agents benefit. The learning network enables organizational knowledge to compound over time.

Recording Learnings

When you discover something useful, record it:

roboco_record_learning(
    content="Use transactions for multi-table updates to prevent partial writes",
    category="pattern",
    team="backend",      # Optional: backend, frontend, ux_ui
    shareable=True,      # Default: True
    tags=["database", "transactions", "consistency"]
)

Learning Categories

Category	When to Use
error_handling	How to handle specific errors
performance	Optimization techniques
testing	Testing strategies and patterns
pattern	Code patterns and idioms
architecture	Design decisions and trade-offs
security	Security best practices
workflow	Process improvements
tooling	Tool usage tips

Searching Learnings

Before starting work, check what others learned:

roboco_search_learnings(
    query="database connection pooling",
    category="performance",  # Optional filter
    team="backend",          # Optional filter
    top_k=10
)

What to Record

DO record:

• Solutions to tricky problems
• Performance optimizations discovered
• Security patterns you implemented
• Testing strategies that worked
• Workflow improvements
• Tool configurations that helped

DON'T record:

• Obvious/basic knowledge
• Temporary workarounds
• Context-specific hacks
• Personal preferences

Good Learning Examples

# Specific and actionable
roboco_record_learning(
    content="Redis SCAN is O(N) total but O(1) per call. Use SCAN over KEYS for large datasets.",
    category="performance",
    tags=["redis", "scan", "keys"]
)

# Pattern with context
roboco_record_learning(
    content="Use circuit breakers for external API calls. Implemented in services/http_client.py:42",
    category="pattern",
    tags=["resilience", "circuit-breaker", "api"]
)

# Security insight
roboco_record_learning(
    content="Always validate file uploads server-side. Client validation is insufficient.",
    category="security",
    tags=["upload", "validation"]
)

Learning Flow

Agent solves problem
        |
        v
roboco_record_learning()
        |
        v
+------------------+
| Indexed in KB    |
+------------------+
        |
        +-- Available via roboco_search_learnings()
        +-- Included in roboco_ask_mentor() responses
        +-- Injected in proactive context for similar tasks
        |
        v
Future agents benefit

Best Practices

1. Record immediately - Don't wait, you'll forget details
2. Be specific - Include file paths, function names
3. Add context - Why does this matter?
4. Tag appropriately - Helps future discovery
5. Search first - Before solving, check if someone already did

Team vs Org Scope

• team filter: Learnings from your cell (backend/frontend/ux_ui)
• No filter: Learnings from entire organization

Cross-team learnings are often valuable - security and performance insights apply everywhere.