[Bug] Excessive token consumption per prompt with Opus 4.7 1M context model

Bug Description
Title: Excessive token usage per prompt — weekly limit exhausted unusually fast Description: Over the past several days, token consumption per single prompt in Claude Code has been abnormally high. Observed: - Weekly usage (all models) hit 100% far earlier than my usual pace - 78% of last-24h usage was at >150k context, even for short follow-ups in fresh sessions - Single prompts that previously cost a few % of the weekly budget now consume significantly more Environment: - Claude Code CLI, Opus 4.7 (1M context) - Plan: Max + extra usage - OS: WSL2 (Ubuntu) on Windows 11 - No recent changes to CLAUDE.md, hooks, MCP servers, or auto-memory Questions: 1. Any recent change to context handling / auto-compact thresholds? 2. Known issue where long sessions fail to compact and re-send full history every turn? 3. Any change in how cached input tokens count against the weekly limit? …
Note: Content was truncated.