The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow to import local files as cloud assets using the import-asset command. This process automatically parses YAML frontmatter from a README.md file located at the provided local path to populate asset metadata. This represents a potential indirect prompt injection surface if the local directory contains attacker-controlled content designed to influence the agent's logic through metadata poisoning.
Ingestion points: The import-asset workflow in SKILL.md reads local-path/README.md frontmatter.
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions when parsing this metadata.
Capability inventory: The skill executes numerous shell commands via the cloudrobo CLI, including asset creation and deletion.
Sanitization: While the skill mentions validation rules for ext_metadata in references/validation-rules.md, these are format checks rather than prompt sanitization.
[COMMAND_EXECUTION]: The skill's primary operation involves the agent executing the cloudrobo asset CLI tool. The instructions provide templates for various subcommands that perform CRUD operations on cloud repositories, catalogs, assets, and versions.
[EXTERNAL_DOWNLOADS]: The installation guide in references/cli-installation-guide.md directs users to download and install the hw-cloudrobo-client package using pip. This is a vendor-provided tool from the skill author (huaweicloud) used to interact with their cloud services.
[DYNAMIC_EXECUTION]: The provided test script scripts/test-cli-commands.sh utilizes python3 -c to execute dynamic code snippets for verifying the Python SDK's functionality. This is a standard practice within the provided testing framework and is restricted to the test environment.

huawei-cloud-cloudrobo-asset