Opus 4.5: excessive confirmation loops and lazy execution requiring repeated prompting

Description

As of late January 2026, Claude Code with claude-opus-4-5-20251101 exhibits a pattern of excessive back-and-forth with the user instead of executing tasks directly. The model frequently:

1. Asks for confirmation instead of acting — even when the instruction is clear and unambiguous
2. Does partial work then stops — requiring the user to re-state the same instruction to get the rest done
3. Spins in circles — appearing to process but producing unnecessary intermediate steps (reading files it already read, asking questions it could resolve itself)
4. Requires repeated prompting — users report needing to tell it the same thing 5-10 times before it fully completes a task

Example Scenario

User: "Add support for model X across all services"

Expected: Claude explores the codebase, identifies all relevant files, and makes all necessary changes in minimal round-trips.

Actual: Claude makes partial changes, stops, waits for user to say "and the other files too", makes one more change, stops again, etc. Each interaction consumes tokens without meaningful progress.

Impact

• Dramatically increases token consumption per task (feels like it's optimizing for token usage rather than task completion)
• Frustrating user experience — feels like fighting the tool instead of collaborating with it
• Tasks that should take 2-3 interactions end up taking 10+
• Users lose trust in the tool's ability to handle multi-file changes autonomously

Expected Behavior

When given a clear instruction like "add X to all services", Claude should:

1. Explore the codebase to find all relevant locations
2. Make ALL necessary changes in one pass
3. Only ask questions when there's genuine ambiguity

Environment

• Claude Code CLI
• Model: claude-opus-4-5-20251101
• Date: January 31, 2026
• Platform: macOS