LLM-Safe Release Context

Use this pattern for release-note, changelog, and report workflows where one node gathers a large source artifact and later LLM nodes plan or write from it.

The key rule: keep the full artifact for audit/debugging, but pass a compact projection to every LLM prompt.

This avoids the common failure mode where a large context.json fills the model window and the workflow fails as "structured output required" before the agent has enough context left to call store-progress.