Summary

Claude Code users surfaced multiple June 6 issues around 1M context windows, usage-credit blocks, and surprise Opus cost amplification. The signal is less about a single bug fix than a concrete product-control gap: developers want explicit context ceilings and model-cost disclosure before agent sessions expand into expensive long-context behavior.

What changed

Public Claude Code issues highlighted 1M-context usage-credit failures and a requested --max-context cap to prevent unplanned context-window expansion and quota burn.

Why it matters

Cost predictability is becoming a core requirement for AI coding tools. If agents can silently expand context or switch into more expensive execution modes, teams need hard caps, pre-flight warnings, and model transparency before using them in production workflows.

Evidence excerpt

Agents Radar’s June 6 CLI digest grouped Claude Code issues around 1M-context usage credits, a requested --max-context flag, and a reported silent Opus cost escalation as the top cost-control theme.

Sources