The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads binary tools such as gitleaks from GitHub and various proxy mirrors (including gh-proxy.com) in scripts/skill_audit.py. It also downloads the skill-quality-cli component from a Huawei Cloud OBS bucket in scripts/install_cli.sh. All external downloads are secured with hardcoded SHA-256 checksums to ensure component integrity regardless of the download source.
[DATA_EXFILTRATION]: A telemetry reporting feature in scripts/cli/cli_reporting.py collects session metadata, user inputs, and step results. It also retrieves historical context from other local agent databases (OpenCode, Hermes, Codex) on the host machine. This information is uploaded to the vendor-controlled domain skillsapi.developer.myhuaweicloud.com for quality monitoring; this behavior is disclosed in the documentation and can be disabled via environment variables.
[COMMAND_EXECUTION]: The tool employs subprocess.run to execute audit tools and the commands it is designed to wrap. It also includes a wrapper script scripts/hcloud-run.sh to ensure all Huawei Cloud CLI (hcloud) calls are audited and reported via the telemetry system.
[DYNAMIC_EXECUTION]: The skill uses os.execv in scripts/cli/cli_entry.py to perform self-updates by replacing the current process with a newly downloaded version. Additionally, the audit logic uses Python's ast module to perform static analysis of Python code to identify security vulnerabilities.
[SAFE]: Automated scanners flagged files like skillspector_rules.json and runtime_security_rules.json for containing malicious patterns such as reverse shells and Base64-encoded execution strings. These are confirmed to be false positives, as these patterns are static regex signatures used for detecting threats in other skills, not executable instructions within this tool.

huawei-cloud-skill-audit