Use ProfileClaw as long-lived context, not prompt history
These guides show how to give agents a stable context layer, keep reads lean, and grow into event-driven or retry-aware orchestration only when the runtime needs it.
Most agents should start with the 2-call path
One context read plus one task-specific call is usually enough to avoid both over-fetching and under-contexting.
Start with a recommended orchestration path
These pages explain how ProfileClaw should sit inside your agent runtime.
Agent Quickstart
Recommended 2-call and 3-call patterns.
Agent Auth
How agents should inject keys, scopes, and failure handling.
Add operational controls only when needed
Caching, retries, and webhooks matter once agents move from experimentation into production loops.
Caching
Keep repeated reads efficient without hiding important changes.
Retries
Retry the right failures with backoff and contract-aware branching.
Agent Webhooks
Move from polling to event-driven continuation when timing matters.