my-skill

The code is primarily an AI-agent sandbox/evaluation runner, not clear malware. It deliberately executes uploaded skill code and dependency installation scripts, exposes configured API credentials to the agent, and disables Claude permission prompts. These are significant operational security risks if the sandbox is escapable, network access is unrestricted, or uploaded skills are untrusted. The shown fragment also contains a syntax error at MONITOR_SCRIPT =. No direct evidence of host compromise, persistence, data theft beyond intended API-key validation, or suspicious exfiltration is present.

The code is a local skill-management utility and shows no evidence of malware, exfiltration, backdoors, or obfuscation. However, user-controlled org and skill_name values can cause path traversal or absolute-path resolution, particularly allowing uninstall_skill() to recursively delete unintended directories and allowing metadata writes or symlink operations outside the intended root. This is a significant input-validation security risk if these arguments can be attacker-controlled.

The code is an execution wrapper for locally installed Python skills. It performs expected dependency synchronization and runs a manifest-defined entrypoint, so using an untrusted skill can execute arbitrary code and potentially install or run untrusted dependencies. No direct malware behavior or obfuscation is evident in this fragment. Review of the imported command-building, manifest-parsing, path-resolution, and environment-building functions is required to determine whether injection or path traversal is possible.

No malicious behavior is evident. The configuration is consistent with local development infrastructure, but it has meaningful security weaknesses if exposed beyond an isolated developer machine: predictable administrative credentials, publicly bound service ports, and non-digest-pinned images. Restrict port bindings, use strong local secrets, and pin images by trusted version and digest for reproducibility.