Contextual, Capability-Driven Primers

Auto-inject small, targeted coaching based on registered capabilities. The system assembles guidance snippets just-in-time so the agent learns how to use relevant tools without polluting context.

Concept

• Triggers fire on registration (e.g., agent.module(conn, name="db")).
• A primer assembler deduplicates, orders, and budgets snippets before each LLM call.
• Snippets are small, composable fragments (e.g., safe patterns for unpicklables, view_image usage).

Benefits

• Reduces boilerplate primers like db_primer.py.
• Keeps context lean; injects only when capabilities are present.
• Teaches best practices at the moment of need.

Considerations

• Budget control integrates with system token counting.
• User-extensible primer library with clear triggers (module/class/instance patterns).

Related issue: Issue #1