Excessive token usage due to startup overhead

Description

Each conversation consumes a significant number of tokens before any user message is processed, due to startup overhead injected into the context window.

Details

On startup, Claude Code injects substantial context including CLAUDE.md files, memory files, system reminders, deferred tool lists, and other metadata. This overhead is charged to the user's token budget on every conversation, even for simple tasks.

Impact

• Users burn through token quotas faster than expected
• The /usage dialog (when visible) may show unexpectedly high consumption
• Cost is non-obvious and hard to predict

Expected Behavior

Startup context overhead should be minimised, or users should be clearly informed of the baseline token cost per session so they can make informed decisions.