CLAUDE AGENTS STALL ON MALWARE SCANS

Users deploying Claude Managed Agents for code generation are encountering a persistent issue where system prompts appended to read operations trigger malware analysis on every file access. The process consumes significant tokens and processing time, increasing operational costs without providing proportional security benefits. After Claude completes the malware scan and confirms files are safe, the appended prompt is still being interpreted as a restriction, causing subagents to refuse legitimate code augmentation tasks. This creates a catch-22 scenario where security measures intended to protect the system actively prevent it from functioning as designed. The issue appears to be a regression—a previously working feature that has broken—affecting developers using managed agents in production repositories. The problem highlights tensions between security safeguards and operational efficiency in AI-assisted development workflows. Users report the malware reminder persists across read operations, suggesting the fix may require adjusting how security prompts are contextualized versus how task restrictions are enforced.