[FEATURE] Native circuit breaker — auto-stop after repeated identical failures

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When a command, fix, or approach fails repeatedly, Claude Code will often retry the same or very similar approach multiple times before giving up — sometimes exhausting the context window. Users currently rely on CLAUDE.md rules like "stop after 3 failures" to prevent this, but the model does not always self-enforce.

Proposed Solution

Implement native loop detection in the agent's execution loop. Track sequential tool calls that produce the same or similar errors. After N failures (configurable, default 3), automatically pause execution and force a user intervention prompt: "This approach has failed 3 times with similar errors. How should I proceed?"

Alternative Solutions

Users add rules to CLAUDE.md telling the model to stop after 3 failures. This is unreliable because it depends on model self-discipline.

Priority

High - Significant impact on productivity

Feature Category

Performance and speed

Use Case Example

Claude tries to fix a TypeScript error. Its fix doesn't work. It tries a similar fix. Same error. It tries again. Same error. By attempt 5-6, the context window is polluted with failed attempts and the model is less likely to find the right fix. A circuit breaker at attempt 3 would save tokens and let the user redirect.

Additional Context

_No response_