Claude pushes context to 1M tokens

This is a real workflow upgrade, not just a spec bump, because long-horizon agent tasks usually fail at context handoffs first.

• –Whole-repo analysis gets more practical for cross-file refactors, dependency tracing, and architecture-level debugging.
• –Large document synthesis improves when teams can keep source material in one working context instead of stitching many chunks.
• –Multi-step agents should stay coherent across longer tool-call chains, especially for coding and research pipelines.
• –Cost and quality tradeoffs still matter at extreme lengths, so teams will need compaction/context hygiene to avoid context rot.