The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The automated detection of a curl output piped to python3 in SKILL.md and references/compute-storage-attacks.md is a false positive. The full command curl -s http://169.254.169.254/openstack/latest/meta_data.json | python3 -m json.tool uses the standard library json.tool module solely for formatting metadata responses into a readable JSON structure and does not execute arbitrary code.
[COMMAND_EXECUTION]: The skill instructs users to install standard cloud management and security assessment packages via pip, specifically python-openstackclient, huaweicloudsdkcore, and huaweicloudsdkiam. These are official or industry-standard tools for interacting with Huawei Cloud and OpenStack-based environments.
[INDIRECT_PROMPT_INJECTION]: The skill defines processes for reading data from external cloud infrastructure endpoints, which serves as a potential surface for untrusted data ingestion.
Ingestion points: Huawei Cloud ECS metadata endpoints (169.254.169.254/openstack/latest/), OBS bucket listings, and various service-specific REST API responses.
Boundary markers: Not explicitly defined within the methodology instructions.
Capability inventory: Extensive use of CLI tools and REST APIs for cloud resource management and data retrieval.
Sanitization: Not implemented, as the skill provides manual testing procedures rather than automated data processing logic.
[SAFE]: All documented techniques, including IAM privilege escalation, bucket enumeration, and container environment exploitation, are presented as methodology for security assessments. The external resources, such as obsutil and official SDKs, originate from official Huawei Cloud domains, consistent with the vendor-focused nature of the skill.

huawei-pentesting