compound-product

This wrapper script is not a standalone malware payload, but it materially increases operational risk by repeatedly invoking an LLM agent/CLI with `--dangerously-skip-permissions` and by logging/printing all merged agent output. The practical threat is that malicious or prompt-injected content in CLAUDE.md (or unsafe claude CLI configuration) could lead to unintended host actions, with outputs potentially exposing sensitive data. Additional context (CLAUDE.md contents, claude CLI defaults, and what the agent is permitted to do) is required to determine whether the overall package behavior is truly dangerous.