context-engineering

Context management prevents hallucinations by coordinating rules, specs, and sources in a strict loading order with enforced boundaries. This ensures the agent only accesses relevant information, reducing drift and maintaining output reliability across multi-step tasks.

Organizing project rules and specs requires context layering, loading project-wide rules first, then per-feature specs, and finally relevant sources. This curated loading sequence establishes consistent boundaries and prompt hygiene for the new session.

Agent drift during task switches happens when context is scattered without central curation. Coordinating context reconfiguration enforces conversation management and error reporting, ensuring consistent guidance and guarding against performance degradation across switches.

Context reconfiguration applies across task switches and active sessions by re-evaluating loaded rules and sources. It enforces conversation boundaries and prompt hygiene to maintain safe, reliable outputs without requiring a full session restart.

Context engineering operates dependency-free, applying internal logic to coordinate rules, specs, and sources. It centralizes context loading natively to improve agent reliability and safety without requiring external frameworks or components.

Avoid automated context layering when a task requires highly dynamic, unpredictable source retrieval outside predefined rules and specs. The skill enforces strict loading order and boundaries, which may restrict flexibility for unstructured, exploratory agent interactions.