The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and installs a CLI utility (skill-quality-cli) from the official Huawei Cloud domain https://skillsapi.developer.myhuaweicloud.com. The installation scripts (scripts/ensure_cli.sh and scripts/install_cli.sh) include SHA256 integrity verification for the downloaded artifacts.
[COMMAND_EXECUTION]: The framework executes shell commands and scripts to perform environment checks, skill installation validation, and resource lifecycle testing. It interacts with the hcloud CLI and various Huawei Cloud SDKs.
[DYNAMIC_EXECUTION]: The skill dynamically extracts Python and Bash code blocks from the SKILL.md files of other skills and executes them at runtime using subprocess.run and Python's exec function. This is the intended behavior for a functional testing framework.
[DATA_EXFILTRATION]: The skill accesses the local Huawei Cloud configuration file at ~/.hcloud/config.json to retrieve AK/SK credentials for API authentication. It follows security best practices by explicitly instructing the agent and user not to share these credentials in the chat interface.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect injection as it ingests and executes content from external, potentially untrusted skill packages.
Ingestion points: Reads SKILL.md, scripts/, and references/ from target skill directories provided by the user or discovered in the install path.
Boundary markers: No specific boundary markers are used to isolate target skill data from execution logic.
Capability inventory: Full shell command execution, Python code execution, and network API access via Huawei Cloud SDKs.
Sanitization: Implements a redaction utility in scripts/lib/utils.sh that uses regular expressions to scrub potential AK/SK patterns and secrets from the generated test reports and logs.

huawei-cloud-skill-tester