The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes large volumes of untrusted data from the Huawei Cloud RMS (Resource Management Service) API. Resource names, descriptions, and other metadata are used to generate Terraform configurations, creating a potential surface for indirect prompt injection if an attacker controls resource properties in the target environment.
Ingestion points: rms_resources_${REGION}.json (output from RMS query).
Boundary markers: No explicit delimiters or instructions are used to separate the external data from the agent's analysis logic.
Capability inventory: The skill executes shell scripts, hcloud (KooCLI) commands, and terraform operations (init, validate, plan).
Sanitization: There is no evidence of sanitization or filtering applied to resource metadata before it is processed by the logic in SKILL.md or the helper scripts.
[DYNAMIC_EXECUTION]: Multiple helper scripts (scripts/query_all_resources.sh, scripts/compare_resources.sh) use shell-based string interpolation to dynamically generate and execute Python code via python3 -c. While used for processing local inventory files, this pattern is susceptible to code injection if file paths or arguments are manipulated.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of external tools from official sources. It downloads the KooCLI via a piped shell script from a Huawei Cloud domain and fetches Terraform from HashiCorp's official release registry. These are recognized vendor and well-known service resources.
Evidence: curl -sSL https://cn-north-4-hwc-cli.obs.cn-north-4.myhuaweicloud.com/cli/install.sh | bash in references/cli-installation-guide.md.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple system-level CLI tools (hcloud, terraform, python3) to perform its discovery and validation workflow. The workflow documentation emphasizes that terraform apply should only be run after manual review, reducing the risk of unintended infrastructure changes.

huawei-cloud-iac-reverse