Agent re-runs tests unnecessarily after context compaction

Description

When a long session triggers context compaction (conversation summary), Claude Code loses track of which verification steps (like test runs) have already been performed. It re-runs the full test suite even when:

1. Tests were already run and confirmed passing moments earlier
2. The results were included in the compacted summary
3. No code changes occurred between the test run and the repeated run

Reproduction

1. Work through multiple sequential code changes in a single session (e.g., refactoring audit with 10+ issues)
2. Each change involves: edit files → run tests → commit → next issue
3. After context compaction occurs mid-session, Claude re-runs the full test suite unnecessarily — sometimes multiple times in a single response — even when tests just passed

Expected Behavior

After context compaction, Claude should trust the test results summarized in the compaction context and not re-run them unless code has changed since the last run.

Impact

• Wastes user time (test suites can take 1-2+ minutes)
• Frustrating UX when the same 100+ second test run happens 2-3 times in a row
• Gets worse as sessions get longer and compaction happens more frequently

Environment

• Claude Code CLI
• Model: claude-opus-4-6
• Long refactoring sessions with 10+ sequential commits