huawei-cloud-skill-audit

This is an LLM-assisted finding triage module, not clearly malware. When enabled with an API key, it transmits source context to a third-party endpoint and lets that service suppress findings. The unsanitized issue path can potentially read files outside the intended directory, and prompt injection may manipulate verdicts. Restrict file paths to skill_dir, review the endpoint and data-sharing policy, and treat remote verdicts as untrusted.

The code is a telemetry/reporting client rather than an overt malware payload. It automatically harvests sensitive AI-agent session data from local files and databases and transmits it to remote endpoints. The arbitrary environment-controlled GUEST_ENDPOINT, lack of redaction, and handling of cloud credentials are the primary security concerns. Use should require explicit consent, strict endpoint validation, data minimization/redaction, and careful credential isolation.

The code appears to implement a telemetry/reporting CLI with self-installation and automatic executable updates. It does not show clear malware such as a reverse shell, credential theft, destructive activity, or mining. However, automatic execution of binaries selected by a remotely controlled endpoint, combined with a server-supplied SHA256 value and environment-configurable endpoints, creates a significant supply-chain risk. Reporting may also disclose sensitive project or execution data depending on the omitted cli_reporting implementation. Independent package signing, opt-in upgrades, endpoint allowlisting, and review of cli_reporting.py are recommended.