huawei-cloud-publish-work-to-gallery
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
scripts/spawn-detached.mjsto launch local projects (e.g.,npm run dev) andscripts/preflight.shorpreflight.ps1to perform system-level installations and environment setup usingyum,apt-get, andpip. - [EXTERNAL_DOWNLOADS]: The skill programmatically downloads external binaries and assets necessary for its operation. This includes fetching the
hcloudCLI binary from Huawei Cloud OBS storage, thegitcode-oauthutility from GitCode, and Noto Color Emoji fonts from Huawei Cloud npm mirrors and jsDelivr. - [CREDENTIALS_UNSAFE]: The skill manages several types of sensitive information: it reads
~/.git-credentialsand uses the Windowscmdkeyutility to detect GitCode authentication tokens, and it usesscripts/gen_sts.pyto generate and save temporary Huawei Cloud IAM STS credentials to local files (sts-creds.json,sts-creds.sh) for platform authentication. - [DYNAMIC_EXECUTION]: The
scripts/spawn-detached.mjsutility allows the agent to execute arbitrary detached processes on the host system usingsetsidon Linux andWin32_Process.Createvia WMI on Windows. This is used to run the user's application locally so it can be screenshotted by the agent. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the user's local project files, such as
README.mdand project manifests. These contents are used to extract information like project names and descriptions, which are then interpolated into API requests and agent prompts.
Audit Metadata